**Optimizing Irradiation Geometry in LED-Based Photoacoustic Imaging with 3D Printed Flexible and Modular Light Delivery System**

**Maju Kuriakose <sup>1</sup> , Christopher D. Nguyen 1, Mithun Kuniyil Ajith Singh <sup>2</sup> and Srivalleesha Mallidi 1,\***


Received: 21 May 2020; Accepted: 29 June 2020; Published: 6 July 2020

**Abstract:** Photoacoustic (PA) imaging–a technique combining the ability of optical imaging to probe functional properties of the tissue and deep structural imaging ability of ultrasound–has gained significant popularity in the past two decades for its utility in several biomedical applications. More recently, light-emitting diodes (LED) are being explored as an alternative to bulky and expensive laser systems used in PA imaging for their portability and low-cost. Due to the large beam divergence of LEDs compared to traditional laser beams, it is imperative to quantify the angular dependence of LED-based illumination and optimize its performance for imaging superficial or deep-seated lesions. A custom-built modular 3-D printed hinge system and tissue-mimicking phantoms with various absorption and scattering properties were used in this study to quantify the angular dependence of LED-based illumination. We also experimentally calculated the source divergence of the pulsed-LED arrays to be 58◦ ± 8◦. Our results from point sources (pencil lead phantom) in non-scattering medium obey the cotangential relationship between the angle of irradiation and maximum PA intensity obtained at various imaging depths, as expected. Strong dependence on the angle of illumination at superficial depths (−5◦/mm at 10 mm) was observed that becomes weaker at intermediate depths (−2.5◦/mm at 20 mm) and negligible at deeper locations (−1.1◦/mm at 30 mm). The results from the tissue-mimicking phantom in scattering media indicate that angles between 30–75◦ could be used for imaging lesions at various depths (12 mm–28 mm) where lower LED illumination angles (closer to being parallel to the imaging plane) are preferable for deep tissue imaging and superficial lesion imaging is possible with higher LED illumination angles (closer to being perpendicular to the imaging plane). Our results can serve as a priori knowledge for the future LED-based PA system designs employed for both preclinical and clinical applications.

**Keywords:** LED; photoacoustic imaging; ultrasound; 3-D printed photoacoustic probe holder; light delivery optimization; LED divergence

#### **1. Introduction**

Photoacoustic (PA) imaging has gained significant popularity for imaging functional and molecular information in both preclinical and clinical settings [1–5]. The technique involves sending light pulses (a few nanosecond pulse-width) into imaging planes that get absorbed by endogenous (e.g., hemoglobin) or exogenous (e.g., Indocyanine Green) tissue chromophores and generate acoustic waves, which can be detected by conventional ultrasound (US) transducers [2,6,7]. Based on the endogenous contrast

provided by hemoglobin, PA imaging has shown promise in vascular functional imaging of human neonatal brains [8,9], malignant lesions [10–14], and monitoring therapies such as photodynamic therapy [1,15,16], etc. As PA imaging uniquely possesses the best properties of optical imaging (high spatial resolution, functional properties, and imaging speed) and US imaging (structural properties and penetration depth reaching tens of cm), its relevance and popularity are continuously increasing in clinical settings [2,4,17–19].

In PA phenomena, the acoustic pressure (**P0**) generated is proportional to the optical absorption coefficient (μ*a*, m−1) of the light absorber and locally available light fluence or radiant exposure (*f0*, Jm<sup>−</sup>2). This can be represented by [2,6,20,21]:

$$\mathbf{P}\_0 = \ \ u\_a f\_0 \tag{1}$$

where, is the dimensionless, material thermal property dependent Gruneisen coefficient.

Light attenuates as it travels down through a material or tissue due to scattering and absorption. Moreover, for a limited aperture illumination, angle of illumination also plays an important role in defining local fluence (Figure 1). As a result, *f0* and thus PA signal intensity, **P0**, changes as a function of depth (distance from transducer or excitation source) and as a function of the illumination angle. Therefore, the optimization of light delivery is crucial for efficient PA imaging and obtaining high signal-to-noise (SNR) ratio at deeper penetration depths [22–26]. Specifically, for reflection mode PA imaging (transducer and light source on the same side of the sample), several studies demonstrated the dependence of irradiation angle and fiber (source)-to-transducer positioning on PA signal at various depths experimentally or through simulations [23,27–31]. Either unilateral or bilateral positioning of fiber bundles aligned with the nanosecond pulsed laser have been employed in these studies. For example, Haisch et al. utilized a mechanical setup that allowed unilateral illumination (one-side of the transducer) with 20–80 degrees range of motion [32]. The fiber bundle aperture size used in that study was shorter than the ultrasound transducer, which may hinder full aperture illumination of near field absorbers and can presumably suitable for intermediate to deep tissue imaging and is more suitable to image smaller lesions on the skin. In another study by Sivasubramanian et al., two fiber bundles were placed on either side of the transducer at a fixed angle. Change in the illumination angle would require changing the holder setup [33]. More recently, Sangha et al. designed a motorized system to change the bilateral light illumination in the 0◦–60◦ range and concluded that the illumination geometry optimization is important to achieve high SNR at different depths [34]. All these studies point out that change in illumination geometry effects PA SNR at different depths and strongly indicate the need for a flexible handheld system that can deliver light at different angles depending on the depth of the lesions.

The light delivery optimization studies mentioned above were performed with spatially low diverging coherent laser sources. Though conventional lasers (e.g., Q-switched optical parameter oscillator (OPO)) can deliver the required pulse energy at various NIR wavelengths, their bulkiness, minimal-portability, and difficulty in operation prevent them from effortless usage in a clinical setting. Interestingly, nanosecond pulsed light-emitting diodes (LED) show promise in being an alternative to lasers, while offering cost-effectiveness and ease of operation has been recently proven to be successful in several studies [35–39]. Despite the low power of LEDs (about 3 orders of magnitude lower than the conventional Q-switched laser sources), their high pulse repetition rate (PRR) (maximum reported up to 16 kHz opposed) gives the opportunity to average several frames in real-time to achieve an SNR on par with conventional laser-based PA imaging (PAI). In addition, the large spatial divergence of LED arrays (~60◦), could aid in irradiating larger sample area and potentially also provide a quasi-uniform illumination over several millimeters without a light diffuser. Given these attributes, LED-based PA systems demonstrate strong potential for clinical translation. As the use of LED's in PA imaging is still in its infancy, it is important to characterize and optimize the light delivery strategies for better SNR at various depths.

**Figure 1.** (**a**–**c**) Schematics of light emitting diode (LED) illumination cross-sectional view of photoacoustic (PA) setup at representative angles: 15◦, 45◦, and 75◦, respectively, orthogonal to the imaging plane *XZ*, that contains a hypothetical absorber, p. M0 is the medium that facilitates acoustic coupling between the transducer and the phantom material M1; (**d**–**f**) Photographs of LED source pivoted at representative angles, θ = 15◦, 45◦, and 75◦, respectively, using 3D printed modular hinge system.

LED array-based PAI studies so far have used a fixed orientation either in the reflection mode [40–42] or transmission mode [38,43]. In this study, we designed a flexible modular light delivery system for reflection mode PAI that is capable of orienting light from the LED arrays at various angles in the range of 0◦–90◦ (Figure 1). Utilizing the flexible light delivery system, we evaluated PA image contrast in various tissue-mimicking phantoms (point targets in non-scattering and scattering liquid media, absorbing lesion under non-scattering liquid and scattering tissue such as the chicken breast) for the PA signal dependency as a function of irradiation angle and depth. We believe that our findings have an important impact on optimizing the design of LED-based PA probes and accelerate its clinical translation towards imaging both deeper and shallower lesions.

#### **2. Materials and Methods**

#### *2.1. Photoacoustic System and Modular Arrangement for Varying Illumination Direction*

#### 2.1.1. AcousticX

An LED-based photoacoustic system (AcousticX from Cyberdyne Inc., Tsukuba, Japan) with linear US transducer (7 MHz central frequency, 128 elements, 0.315 mm pitch, and 38.4 mm aperture size, elevation focus of 15 mm) and two 850 nm LED arrays (30 to 150 ns pulse width, 4 kHz maximum repetition rate, 200 μJ pulse energy for each array, 5 mm × 40 mm aperture size, 60◦ divergence) on both sides of the US detector was used for the experiments [44]. PA and US raw data were sampled at 40 MHz and 20 MHz, respectively, and data was reconstructed in real-time using an inbuilt Fourier-domain reconstruction algorithm of the system. For offline analysis, both PA and US data were reconstructed using a previously reported frequency domain beamforming algorithm [45]. Radiant exposure per pulse at the LED array surface is about 100 μJ/cm<sup>2</sup> (200 μJ/pulse in an array area of 2 cm2). Given the LED source divergence of 60◦ and ~10.5 mm distance between the US transducer and the phantom surface to accommodate LED arrays for different angles, maximum radiant exposure at the phantom

surface was estimated to be about 29 μJ cm−<sup>2</sup> (for an area of about 6.93 cm2) from a single array at 0◦ LED illumination angle.

#### 2.1.2. Flexible LED Holder: Modular Design for Adjusting Irradiation Direction

Two identical modular hinge systems were designed, 3D printed and used to pivot LED arrays at different illumination directions with respect to the imaging plane. The modular LED holder consisted of three parts that were designed on Autodesk and printed using polylactic acid (PLA) on a MakerBot system. All pieces were joined together with 8–32 socket head screws. (Figure 2). Each of the modules consisted of four hinges that were attached to one another. These hinges can be adjusted or pivoted to create the required angle of illumination. Modules were attached to the US transducer on its one end and the LED arrays were gripped through the heat sink of the arrays on the pivoting end of the module, as shown in the figure. The inter LED array distance (between their adjacent edges) was about 1 cm to accommodate the US transducer. Experiments were done for 0◦, 15◦, 30◦, 45◦, 60◦, 75◦, and 90◦ angles using this modular arrangement. The LED array angles were adjusted with respect to the central axis of the US transducer using a custom-made protractor as shown in Figure 1d. During the experiments, both the US transducer and sample position were unaltered and only the LED sources were adjusted, to avoid PA intensity variations due to sample motion with respect to the US transducer.

**Figure 2.** (**a**) Modular hinge system holding the LED arrays and attached to the US transducer. (**b**) The photographs of the individual pieces are shown in the top panel. The 3D renderings of the hinge pieces are shown on the bottom panel. Part A fits around the transducer and extends the horizontal reach of the holder to allow the LEDs to be placed at angles approaching 90◦. Part B in conjunction with Part A allows precise horizontal and vertical height adjustment. The holder consists of two-part B pieces, and schematic of only one piece is shown in the panel. Part C holds the LEDs using the heat sinks and provides flexibility for any final adjustments on the LED illumination angle. Scale bar = 10 mm.

#### *2.2. Phantoms*

#### 2.2.1. Graphite Pencil Lead Phantoms

A matrix of pencil leads (Graphite 2B 0.5 mm manufactured by June Gold, Bountiful, UT USA), arranged in 4 rows × 5 columns with a spacing of about 5 mm (columns) and 6 mm (rows), was constructed using two 3D printed plastic holders as shown in Figure 3a. The phantom construction with pencil lead is immersed in a container with water or 1% Intralipid (Sigma Aldrich Inc., Atlanta, GA, USA) solution (Figure 3b). The scattering coefficient of 1% intralipid solution is 1.8 mm−<sup>1</sup> [46,47] close to the values reported for tumor tissue (1–2 mm<sup>−</sup>1) [48–50]. All experiments were conducted at room temperature (22 ◦C).

**Figure 3.** (**a**) Photograph of Pencil lead matrix; (**b**) photograph of an experimental arrangement using intralipid medium; (**c**–**h**) PA image acquired at representative angles 15◦ (**c**,**f**), 45◦ (**d**,**g**), and 75◦ (**e**,**h**) in water (**c**–**e**) and 1% intralipid (**f**–**h**); (**i**,**k**) Mean PA signal intensities and their standard deviations (of 5 lateral positions at each depth) plotted as a function of LED angles in water (**j**) and intralipid (**k**); and (**j**,**l**) corresponding contrast to noise ratios (CNRs) obtained as a function of LED angles in water (**j**) and intralipid (**l**). Different depths from the transducer are indicated by 12 mm (line with black circles), 18 mm (line with red squares), 24 mm (line with blue diamonds), and 30 mm (line with green downward triangles). The noise background levels corresponding to 12, 18, 24, and 30 mm are represented by black, red, blue, and green dash-dotted lines, respectively, in (**i**) in water and (**j**) in intralipid. The backgrounds were obtained right below from each signal regions, e.g., as indicated by the yellow rectangles in (**c**,**f**). Images for all the angles can be found in Figure S2 (for water) and Figure S3 (for intralipid).

2.2.2. Tissue Mimicking Phantom Containing Lesion with High Optical Absorbance

Tissue mimicking phantoms were prepared using agar powder (Sigma-Aldrich Inc., Atlanta, GA, USA). Titanium (IV) oxide, anatase powder (99.8%, Sigma-Aldrich Inc., Atlanta, GA, USA) was added to provide acoustic contrast and enhance the optical scattering properties of the agar. The preparation was done by slowly adding 1% wt./vol. of agar powder and 1% wt./vol. of TiO2 powder into continually stirred deionized water at ambient conditions to avoid clumps. The final solution was then heated above 80◦ C, above the melting temperature of agar, and exposed to a vacuum level of about 0.1 atm for 5 min to degas the solution and cooled it down to room temperature to obtain the final phantom. A cylindrical light-absorbing lesion with acoustic scatterers was prepared in a similar aforementioned method. Additionally, 0.5% wt./vol. graphite powder (<20 μm, synthetic graphite, Sigma Aldrich Inc., Atlanta, GA, USA) and 0.5% wt./vol. TiO2 powder were added. The concentration of the absorbing and scattering particles was chosen to mimic tumor tissue with an absorption coefficient (~0.2 cm<sup>−</sup>1) and reduced scattering coefficient (~10 cm−<sup>1</sup> ) as previously reported in the literature [46–51].

#### *2.3. Signal Analysis*

#### 2.3.1. PA Intensity & Contrast to Noise Ratio (CNR) Calculation

The PA signal intensity of each pencil lead in the phantom was calculated by taking the maximum pixel value from the region of interest (ROI) around the target (white rectangle in Figure 3c). All the five laterally positioned PA intensities were then averaged to find mean and standard deviation (σ) of PA intensities corresponding to each depth location and illumination angles. Background (Bg) was calculated from an ROI below each point target (yellow rectangle in Figure 3c). It is important to note that the noise/background values were calculated from regions close to the signal ROIs. We chose the ROIs within close proximity of the signal ROI and not from regions at the corner or with only electronic noise, including reconstruction related artifacts that may be present when changing the illumination angle. To plot PA intensity changes as a function of angle in the tissue-mimicking phantom, PA signal intensities from the lesion was obtained by choosing the median PA intensities above the Bg level. Here also, similar to the pencil phantom case, Bg and σ were chosen from a region close to the lesion ROI. PA intensity was plotted in decibel (dB) with the formula:

$$PA\text{ }intensity\text{ }in\,dB\text{ }=\,10\log\_{10}(PA\text{ }intensity)\tag{2}$$

and the CNR was calculated using the formula:

$$\text{CNR in dB} = 10 \log\_{10} \left( \frac{\text{PA intensity} - \text{Bg}}{\sigma} \right) \tag{3}$$

#### 2.3.2. Divergence of the LED Source

Divergence of the LED source is an angular measure of the increase in irradiation area (and corresponding diameter or radius) with distance from the source. Laser light sources are known to have very low divergence while LED sources have high divergence. Sources with high divergence have lower radiant exposure per unit area on the target at a given depth than sources with lower divergence. These changes in radiant exposure can influence PA signal and hence it is critical to evaluate the divergence of the source and choose appropriate illumination angle to obtain maximum CNR. Assuming the LED is a line source aligned in the *X*-axis (parallel to the US transducer) while the emitted wavefronts take quasi-cylindrical shape in the imaging volume, an approximate divergence of the source in the *YZ* plane (Figure 4), in degrees can be computed using:

$$\text{Divergence angle} = 2 \times \left\{ \tan^{-1} \left( \frac{\text{target depth}}{\text{LED to detector distance}} \right) \right\} \tag{4}$$

**Figure 4.** (**a**) Illustration of the source wavefront profile, while assuming the LED array as a line source with Gaussian profile having divergence described by Equation (4); (**b**) Source profile after pivoting LED arrays at three different angles in the imaging plane–green at 0◦, red at 45◦, and blue at 90◦; (**c**) Normalized PA intensity obtained from 18 mm target vs. LED illumination angle, θ, is indicated in blue squares. The red solid curve shows the Gaussian fit using Equation (5), with an R-squared value of 0.95; (**d**) Blue squares with error bar show the peak PA intensity with standard deviation for each depth plotted as a function of θ for pencil lead phantom data in water. The black solid curve displays the best fit using a cotangent function (R2 = 0.95).

The multiplying factor 2 is used to include both sides of the illumination plane. Target depth and LED to detector distance refer to the absorber (pencil lead) position in depth below the detector (US transducer) and its lateral distance to the source (LED array), respectively. In our experiment, the target and LED to detector distance were fixed while the source was pivoted, as shown in Figure 4b. We assume that the LED source exhibits a Gaussian spatial intensity profile and thus a Gaussian function is used to fit PA intensity vs. LED- illumination angle data to find the divergence of the LED arrays, given by

$$f(\mathbf{x}) := \operatorname{a.e.}^{-\left(\frac{\mathbf{x} - \mathbf{b}}{c}\right)^2} \tag{5}$$

where *x* is the source angle, *a* is the peak PA intensity and that corresponds to angle *b*, and *c* is the half of the angle span at which PA intensity shows half maximum (50%) or −6 dB roll-off. The divergence can then be calculated by multiplying c by two, resulting in the full width of the angle at half maximum of the PA intensity (FWHM). The model can be fitted using a least-squares minimization method to obtain best-fit parameters. Moreover, the position of the peak intensity for a chosen depth can be described by

$$d = m + n \cdot \cot(\theta) \tag{6}$$

where θ is the LED illumination angle (Figure 1), *m* is the offset of imaging plane in depth, *n* is the separation between LED and detector, and *d* is the imaging depth.

#### **3. Results**

Experiments were conducted in two different phantom environments: liquid and chicken tissue. The first phantom consisted of pencil lead (point source) as absorbing targets (Figure 3) in water or 1% intralipid media. The second tissue-mimicking phantom consisted of an absorbing cylindrical lesion made of graphite powder (Figure 5), which was placed on top of a chicken breast tissue arranged obliquely to the imaging plane, while the top layer was interchanged between water or chicken breast tissue. The experiments were designed to probe the phantom at all depths simultaneously for each angle of choice, thereby reducing experimental uncertainties.

**Figure 5.** (**a**) Schematic of the tissue-mimicking phantom: Absorbing lesion (cylindrical in shape, 2 mm in diameter) placed on chicken tissue and arranged obliquely in the *XZ* plane. (**b**) Photograph of the experiment. The dotted line indicated by arrowhead shows the projection of lesion to the *x*-axis; (**c**,**d**) US image of the sample showing water top layer (**c**) or chicken layer (**d**) lesion and bottom chicken tissue in both cases; (**e**–**j**) PA intensity images captured using water as the top layer (**e**–**g**) or chicken tissue as the top layer (**h**–**j**), by choosing LED array directions: 15◦ (**e**,**h**), 45◦ (**f**,**i**), and 75◦ (**g**,**j**) with respect to the imaging plane.

#### *3.1. Pencil Lead Phantom Experiments*

3.1.1. Pencil Lead in Scattering and Non-Scattering Media Shows Weak Dependency on the LED Illumination Angle

Initially, PA intensities were monitored as a function of the illumination angle using a 4 × 5 pencil-lead matrix (as detailed previously) aligned orthogonal to the imaging axis to form point targets at defined locations. LED directions were varied from 0◦ to 90◦ in steps of 15◦, as shown in Figure 1. Figure 3a,b shows the photographs of the lead matrix and an experimental arrangement in intralipid, respectively. PA intensity images (in dB scale) obtained at three representative angles: 15◦, 45◦, and 75◦

in water and 1% intralipid medium are shown in Figure 3c–e, Figure 3f–h, respectively. PA images from all the angles between 0◦ and 90◦ can be found in Figure S2 (for water) and Figure S3 (for intralipid). Angle dependent PA intensities and CNR were calculated for four depths (approximately at 12 mm, 18 mm, 24 mm, and 30 mm) as described in Section 2.3 and plotted in Figure 3i, Figure 3k, Figure 3j, Figure 3l, respectively. An approximate gap of 10.5 mm from the US transducer to the sample was kept avoiding near field reconstruction errors and also to allow room for LED adjustments. For non-scattering media (water), the PA signal intensity increased as LED illumination angle increased from parallel orientation (0◦), and then reached a maximum value at intermediate angles and finally decreased in its strength as the angle approached 90◦ (LED illumination perpendicular to the transducer). It is obvious from the plots that the angles corresponding to the maximum PA intensity value were in an inverse relation to the target depth, by showing a maximum value at a steeper angle for the 12 mm target and a maximum value at a shallower angle corresponding for the 30 mm target.

Figure 3j demonstrates CNR obtained in water as a function of the angle of irradiation. It reveals that the PA contrast did not change significantly after reaching a maximum level for an extended angle range of 15◦ to 75◦. This was due to an increase in the background (due to various artifacts) along with the target signal increase that in turn reduced the angle dependency. PA intensities using intralipid (Figure 3k,l) showed a similar trend as water, especially for angles between 15◦ to 75◦. In the cases of 0◦ and 90◦, even though the trend was similar to that of the water phantom, the changes in PA intensity were less for intralipid phantom as expected. This reduction in intensity is due to reduced fluence due to optical scattering that reduces the incoming light directionality (and thus the angle dependency) as opposed to the case of non-scattering water medium. Comparing the depth-dependent CNR within the quasi angle-independent regime (15◦–75◦), the total drop was larger in scattering media (~30 dB) than in the water phantom (~15 dB). This can be due to a larger attenuation promoted by increased light scattering.

#### 3.1.2. LED Source Divergence and Optimum Illumination Angle from Pencil Lead Targets in Water

Knowing the target location in depth and its lateral separation from the illumination plane, the source divergence (Figure 4) orthogonal to the illumination plane (*Y*) can be computed. In our experiments, the center of the light source (approximated here as a line source) was located at 10.5 ± 2 mm away from the imaging plane (imaging axis of the US transducer). Utilizing Equation (5), the divergence of the LED source was calculated to be 58 ± 8◦ at FWHM from the Gaussian fit of PA intensities from pencil lead target at 18 mm depth (Figure 4c). The experimentally derived divergence value is in good agreement with the manufacturer's data (60◦, from Cyberdyne Inc., Tsukuba, Japan). The peak PA intensity (coefficient *b* in Equation (5)) at 18 mm depth was observed at 44◦ ± 3◦. We further analyzed the data in Figure 3i to infer the angle at which maximum PA intensity could be obtained as a function of depth (Figure 4d) using the coefficient *b* in Equation (5). The black solid curve in Figure 4d shows the cotangent fit (Equation (6)) to the data shown in blue squares with a goodness of fit (R-squared) value equal to 0.95. Fitting was done by choosing *m* (depth offset) and *n* (separation of LED to the detector) as free fit parameters. Best fits (and 95% confidence interval bounds) obtained for, *m* is 9 mm (−0.6, 18.6 mm) and *n* is 10.7 mm (3, 18 mm). A large offset value might be due to the experimental error coming from the spatial width of the LED array, which takes up ~10.5 mm below the US transducer, in the imaging plane, at its steepest angle (90◦). The value of *n* matched well to our experimentally set approximate value of 10.5 mm. From the fit, maximum PA signal intensity value at 12 mm can be obtained with a LED illumination angle of 74◦ while 27◦ can be used to image lesions at 30 mm depth. It should be noted that there exists a strong dependency of the illumination angle at superficial regions (slope of −5◦/mm at 10 mm), which weakens as it goes to deeper locations (−2.5◦/mm at 20 mm and −1.1◦/mm at 30 mm) due to the nature of the cotangent function. The diameter of the projected beam onto the imaging plane, *Z*, at a given θ is estimated to be about 11.2 mm (θ = 84.5◦) and 11.8 mm (θ = 27◦) at 10 mm and 30 mm depths, respectively. Interestingly, less than 4% variation in the beam diameter in the imaging range is observed with different LED illumination angle. So,

the large source divergence of about 11–12 mm overcomes the strong angle dependencies and can be used to lower the number of illumination angles for imaging lesions at various depths as is the case with Laser illumination. The LED illumination angle effects are even less notable in scattering medium, where deep tissue imaging can be achieved with smaller LED illumination angles (closer to being parallel to the imaging plane) and superficial lesion imaging is possible with larger LED illumination angles (closer to being perpendicular to the imaging plane).

#### *3.2. E*ff*ect of Surrounding Media and Illumination Direction on PA Signal from Tumor Mimicking Lesion*

A second set of experiments were conducted using tumor mimicking light-absorbing lesion placed obliquely in the imaging plane on top of chicken breast tissue (backing layer) while using water (Figure 5a) or scattering chicken tissue as the top layer (Figure 5b). This phantom study aimed to simulate superficial or deep-seated tumor lesions filled with blood vessels or contrast agents. US images corresponding to experiments using a top water layer (Figure 5c) and top chicken layer (Figure 5d) show the lesion placement and surrounding layers for comparison with corresponding PA images. The PA images using water as the top layer or chicken tissue as the top layer at LED source angles 15◦, 45◦, and 75◦ are shown in Figure 5e–g, Figure 5h–j, respectively. An apparent lateral shift in lesion positions between the images generated with water and chicken top layers was due to a change in imaging transducer placement about 2 mm in the horizontal axis, between the experiments (Figure 5h–j), which had negligible or no effects in our depth-dependent analysis.

Experiments with the top water layer show PA signals from the lesion for the entire imaging depth (Figure 5e–g), which can be associated with the negligible light scattering in water, opposed to chicken tissue (Figure 5h–j). Angle dependent PA intensity variations were also visible in both cases, where deeper tissue illumination was achieved at lower incident angles while increased PA intensity in the detector vicinity was observed for higher illumination angles. These results are very similar to those observed in Figure 3 using pencil lead phantom in water and intralipid. For a detailed analysis of depth-dependent PA intensity variation, Figure 6 was presented with analyses from three depth locations at 12 mm, 20 mm, and 28 mm, in which PA signals were plotted as a function of illumination angle, for the tissue-mimicking phantom with the top water and chicken layers. The white parallelograms in Figure 6a indicate the selected ROI from where the median PA intensity was calculated and the yellow parallelogram ROIs were considered for the background. Calculated PA intensities and CNRs as a function of angle for different depths are shown, respectively, in Figure 6b,c while using water as the top layer, and Figure 6d,f for chicken breast as the top layer. In the case of water as the top layer, the illumination angle parallel to the imaging plane, i.e., 0◦, produced highest PA intensity (~23 dB) at the bottom (28 mm) than at the top regions; at 12 mm, the intensity dropped to ~17 dB as demonstrated in Figure 6b. On the other hand, experiment with the chicken top layer (Figure 6d) showed almost equal but low PA intensity (~17 dB) for all depths at 0◦. It is interesting to note that the PA signal from 12 mm was quasi-constant for all angles from 30◦ and above. For the intermediate depth (20 mm), the intensities showed an increase up to 15◦, then stayed almost unchanging until 60◦ and showed a slight decrease in the mean value with further increase in angle from 75◦ and above. It should also be noted that the signal from 28 mm depth while using chicken top layer was very close to the background level (CNR is less than 5 dB), which shows the maximum penetration depth achieved within our experimental limits.

**Figure 6.** (**a**) PA image of a sample containing absorbing lesion (2 mm diameter) with the top water layer. White parallelograms indicate the regions selected for PA intensity and yellow for the background; (**b**,**d**) PA signal intensity plotted as a function of LED angles for selected depths in mm as labeled by numbers with the corresponding color, right next to each plot. Black circle with line corresponds to 12 mm, red squares with line corresponds to 20 mm and blue diamond with line indicates 28 mm. The black dotted lines, red dash-dotted lines, and blue dashed lines represent the background levels for 12, 20, and 28 mm depths (from the transducer) respectively; a gap of 10.5 mm exists between the transducer and the phantom surface. (**c**) CNR of the lesion under water; (**e**) CNR of the lesion under chicken breast. The error bars represent the standard deviation.

#### **4. Discussion**

The dependence of image contrast on irradiation parameters such as the angle of irradiation relative to the transducer, wavelength of the light irradiation, and distance between the transducer and the light source is undisputed. In this paper, we studied the dependence of signal intensity and contrast in PA images generated by an LED light source at various illumination angles. The 3D modular hinge design gave extreme flexibility to adjust the illumination angle as well as the transducer to LED-array distance. In the current work, the distance between the LED light source and transducer was relatively constant with the LED array positioned very close to the transducer (Figure 1). Studies addressing different separation distances between the transducer and the LED array can potentially give complementary information to our results. With respect to the 3D modular hinge system itself (Figure 2), several design criteria could be improved. Currently, the footprint of the modular system is large (11.5 cm at its widest dimension). Though the complete probe is lightweight due to lack of any heavy motors or metallic pieces, it is still larger than the other 3D printed fixed angle holders, e.g., by Sivasubramanian et al. [33]. The hinge system can be further modified with computer-controlled micro-hinges, can be adapted to any fiber bundles irrespective of their shape or size and can also be integrated with any laser or pulsed diode-based systems.

In all our results, it is evident that the background signal is also increasing along with the PA signal of interest (Figure S1), resulting in less impact on CNR when the illumination angle is increased beyond a certain limit. We believe that the background signal may be affected by various artifacts like reflection artifacts, out-of-plane clutter, and side lobes. For example, it is well known that more reflection artifacts are caused by high PA signal from tissue/phantom surface reflecting off acoustically dense structures when the illumination angle is steep and the fluency is high just beneath the US probe. These reverberation type of reflection artifacts are visible in our results too (Figure 3 and Figure S1) and would have impacted the CNR calculation. At lower illumination angle setups, light can scatter outside the imaging plane, get absorbed by different features, and generate out-of-plane artifacts and this also may have an impact in the CNR [52]. We strongly believe that it is also important to consider these common artifacts in the CNR analysis hence we chose a region very close to the target of interest for the calculations, instead of choosing a blank image (image generated with no light) or electronic noise. Furthermore, we used 850 nm irradiation, a wavelength at which there is relatively high penetration

depth in tissue. The utility of other illumination wavelengths will impact the PA signal intensity and CNR based on the absorption properties of lesions at those wavelengths. Our future studies will involve evaluating these observations especially in an in vivo situation with both subcutaneous (superficial lesions) and orthotopic tumors (deep lesions).

A recent study by Agarwal et al. demonstrated that single-shot laser-based PAI and LED-based PAI achieved the same SNR from lesions 2–3 cm deep in chicken tissue. Though high frame averaging (2560 frames) was performed in LED-based PAI, they achieved real-time imaging due to the high pulse repetition frequency of LED sources [53]. The utility of LED-based PAI systems has been extensively reviewed by Zhu et al. [37] and one of the key factors currently hindering the clinical translation of LED-based photoacoustic imaging systems regardless of demonstrated promise in multiple preclinical and clinical imaging applications is the optical output power of the LED arrays [37,54]. It is of paramount importance to improve the optical output power of LEDs to enhance its usage in a wide range of deep-tissue imaging applications, and thus accelerate the clinical translation. The pulse repetition rate of LEDs is several-fold higher than pulsed LASER systems and it is feasible to average multiple image frames to improve SNR without compromising on real-time imaging capability. However, averaging N frames can improve SNR by only <sup>√</sup>N, and thus this approach has its limitations. Recently several developments in beamforming methodologies are made to improve the CNR and SNR of the LED-based PA systems [55,56]. We believe that the improvement of optical pulse energy and the development of novel image reconstruction and enhancement algorithms will be critical to accelerate the clinical translation of LED-based PAI [54].

#### **5. Conclusions**

Our results show that the optical excitation using an LED source behaves differently than the laser excitation due to a large source divergence of LED arrays, which we calculated to be 58◦ ± 8◦. This in turn reduces the source direction dependency of PA signal at different depths. Our analysis in the non-scattering medium shows a strong dependence of illumination angle vs. depth at near field regions (−5◦/mm at ~10 mm) and weak dependence at deeper locations (−2.5◦/mm at 20 mm; −1.1◦/mm at 30 mm). On the other hand, results from tissue-mimicking phantom in scattering media showed significantly weaker angle dependence of PA signal intensity than the phantom in water. So, utilizing an LED-based system (or a source with similar divergence) for either deep tissue lesions or superficial lesions would be less cumbersome in terms of source alignment by offering the freedom to choose a wide range of irradiation angles without losing CNR in the images. In contrast, spatially coherent sources would require stringent alignment strategies for illuminating lesions at various depths [57,58]. The modular light delivery system and results presented in this study can serve as a priori knowledge for future LED-based PA system designs and aid in further catapulting its utility in both preclinical and clinical applications.

**Supplementary Materials:** The following are available online at http://www.mdpi.com/1424-8220/20/13/3789/s1, Figure S1: (**a**) A photograph of Pencil lead array placed in water bath in the presence of PA transducer on top; (**b**) Schematic of the pencil lead array; (**c**–**h**) PA image acquired at representative angles 15◦ (**c**,**f**), 45◦ (**d**,**g**), and 75◦ (**e**,**h**) in water (**c**–**e**) and 1% intralipid (**f**–**h**); Differ to Figure 2 in the main article, where the reconstructed images are shown in a dynamic range of 30 dB, these images are shown in 40 dB to show the presence of background. Figure S2: Pencil lead in water at different angles of illumination mentioned as in each image title. PA intensity values are given in the bottom color bar. Figure S3: Pencil lead in intralipid at different angles of illumination mentioned as in each image title. PA intensity values are given in the bottom color bar.

**Author Contributions:** Conceptualization, M.K., M.K.A.S., and S.M.; methodology, M.K., C.D.N., and S.M.; software, M.K.; validation, M.K. and C.D.N.; formal analysis, M.K.; investigation, M.K.; resources, S.M.; data curation, M.K. and C.D.N.; writing—original draft preparation, M.K. and S.M.; writing—review and editing, M.K., M.K.A.S., C.D.N., and S.M.; visualization, M.K. and S.M.; supervision, S.M.; project administration, S.M.; funding acquisition, S.M. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was funded by the School of Engineering, Tufts University. Partial salary support of M.K and S.M. by the National Institute of Health RO1CA231606 grant is gratefully acknowledged.

**Acknowledgments:** The authors would like to thank Michael D. Kennedy, Department of Biomedical Engineering, Tufts University, Medford, MA, USA, for 3D printing the modular holder for LED arrays.

**Conflicts of Interest:** M.K.A.S. is employed by CYBERDYNE INC. The authors have no financial interests or conflicts of interest to disclose.

#### **References**


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

## *Article* **Photoacoustic Imaging of Human Vasculature Using LED versus Laser Illumination: A Comparison Study on Tissue Phantoms and In Vivo Humans**

**Sumit Agrawal 1,† , Mithun Kuniyil Ajith Singh 2,† , Kerrick Johnstonbaugh 1, David C. Han 3,4, Colette R. Pameijer <sup>4</sup> and Sri-Rajasekhar Kothapalli 1,5,6,\***


**Abstract:** Vascular diseases are becoming an epidemic with an increasing aging population and increases in obesity and type II diabetes. Point-of-care (POC) diagnosis and monitoring of vascular diseases is an unmet medical need. Photoacoustic imaging (PAI) provides label-free multiparametric information of deep vasculature based on strong absorption of light photons by hemoglobin molecules. However, conventional PAI systems use bulky nanosecond lasers which hinders POC applications. Recently, light-emitting diodes (LEDs) have emerged as cost-effective and portable optical sources for the PAI of living subjects. However, state-of-art LED arrays carry significantly lower optical energy (<0.5 mJ/pulse) and high pulse repetition frequencies (PRFs) (4 KHz) compared to the high-power laser sources (100 mJ/pulse) with low PRFs of 10 Hz. Given these tradeoffs between portability, cost, optical energy and frame rate, this work systematically studies the deep tissue PAI performance of LED and laser illuminations to help select a suitable source for a given biomedical application. To draw a fair comparison, we developed a fiberoptic array that delivers laser illumination similar to the LED array and uses the same ultrasound transducer and data acquisition platform for PAI with these two illuminations. Several controlled studies on tissue phantoms demonstrated that portable LED arrays with high frame averaging show higher signal-to-noise ratios (SNRs) of up to 30 mm depth, and the high-energy laser source was found to be more effective for imaging depths greater than 30 mm at similar frame rates. Label-free in vivo imaging of human hand vasculature studies further confirmed that the vascular contrast from LED-PAI is similar to laser-PAI for up to 2 cm depths. Therefore, LED-PAI systems have strong potential to be a mobile health care technology for diagnosing vascular diseases such as peripheral arterial disease and stroke in POC and resource poor settings.

**Keywords:** deep tissue imaging; hemangioma; laser; light-emitting diodes (LED); mobile health; peripheral arterial disease; photoacoustic imaging; stroke; vascular malformations

#### **1. Introduction**

Vascular diseases are the leading cause of death worldwide. Some common vascular diseases include cardiovascular disease, stroke and peripheral artery disease (PAD) [1–3]. Many of these vascular diseases need point-of-care (POC) diagnosis and monitoring using

**Citation:** Agrawal, S.; Kuniyil Ajith Singh, M.; Johnstonbaugh, K.; C. Han, D.; R. Pameijer, C.; Kothapalli, S.-R. Photoacoustic Imaging of Human Vasculature Using LED versus Laser Illumination: A Comparison Study on Tissue Phantoms and In Vivo Humans. *Sensors* **2021**, *21*, 424. https://doi.org/10.3390/s21020424

Received: 16 December 2020 Accepted: 6 January 2021 Published: 9 January 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

nonionizing, noninvasive and cost-effective approaches. Although Doppler ultrasound meets all these requirements, it only maps blood flow, which is operator dependent and influenced by motion artifacts, resulting in limited sensitivity and specificity to detect the disease in its early stage [4]. A POC technique that provides direct label-free molecular and functional information of vasculature is needed to reliably detect and monitor vascular diseases [5,6].

Photoacoustic imaging (PAI) is a hybrid imaging modality that provides rich optical spectroscopic contrast at ultrasonic penetration depths and resolutions [7]. Over the past two decades, PAI has emerged as a promising tool for label-free imaging of individual blood vessels [8,9], detection of angiogenesis [10] and has also helped in extracting several physiologically relevant parameters such as blood oxygen saturation [11,12] and changes in the blood volume [13], which are vital for monitoring disease progression [14,15]. The basic mechanism of PAI includes a nanosecond pulsed optical excitation that illuminates the biological subject. The light absorbing molecules inside the tissue undergo thermoelastic expansion and generate broadband acoustic waves which are subsequently detected using conventional ultrasound (US) detectors [7]. Biomedical PAI mainly capitalizes on the intrinsic absorbers present in human tissue [8], such as oxy- and deoxyhemoglobin [10,11], melanin [16], lipids [17,18], water [19], RNA and DNA [20], as each of these exhibits a characteristic absorption spectrum. However, if the spectral contrast from these intrinsic chromophores is not sufficient to reveal the disease, a wide range of extrinsic contrast agents [21–25] can be functionalized to target different diseased biomarkers to increase molecular sensitivity and specificity.

While ultrasound and certain optical technologies are available in the size of a mobile phone [26,27], PAI systems still have to reach that level of portability. Conventional PAI systems employ bulky class-IV laser sources (100 mJ/pulse) and data acquisition systems to increase the peak imaging performance [28–31]. Such lasers not only increase the system cost and footprint but also carry a high risk of class-IV exposure. However, to translate PAI technology to POC clinical applications and to resource-limited settings, a significant reduction in both cost and size is required. To address this challenge, several cost-effective alternatives for both the optical excitation [32–35] and the ultrasound detection [36–39] components have been explored, including for wearable applications [40,41].

Recently, low-power laser diodes [42,43] and light-emitting-diodes (LEDs) [44] have been proposed as alternatives to laser sources. Specifically, pertaining to their noncoherent nature, LEDs carry a huge potential to be the safe and cost-effective alternative illumination sources for PAI [43,44]. However, compared to the class-IV lasers, LED arrays carry much less optical energy (i.e., order of 100- s μJ) per pulse, and thus their PAI capabilities as a function of imaging depth need to be studied in detail to employ a suitable optical source for a given POC application [45,46].

To enhance the performance of LEDs, an arrayed arrangement of LED elements was developed [47,48], thereby increasing the pulse energies from a few μJ to hundreds of μJ. In addition to this, higher pulse repetition frequency (PRF) rates (i.e., ~4 KHz) of the LEDs allowed a sufficient PA frame averaging which led to significant signal-to-noise ratio (SNR) improvements for deep tissue targets [46]. These LED array B-mode PAUS [47,48] and tomographic imaging [49,50] setups have been demonstrated using several preclinical small animals [51,52] and in vivo human imaging studies [53–55].

To date, there is no study that quantitatively compares the PAI performance of LEDs and laser illumination head-to-head. Given these tradeoffs between portability, cost, optical energy and frame rate, this work systematically studies the deep tissue PAI performance of LED and laser excitation to help select a suitable source for a given biomedical application. First, a setup for sequentially performing PAI with these two optical illuminations has been developed. Controlled studies on different tissue phantoms have been performed for detailed evaluation of the imaging performance. Further, in vivo human hand vasculature imaging in 2-D and 3-D was performed using these two optical sources. The rest of the paper is organized as follows. Section 2 describes our proposed setup for comparing the

two sources. Comparison studies including tissue-mimicking phantoms and the in vivo human imaging are presented in Section 3. Section 4 provides a detailed discussion of the results.

#### **2. Materials and Methods**

In this section, a detailed description of the experimental setup for studying the PAI capabilities of high-power laser and LED array sources is presented. First, the commercial LED array-based B-mode PA and ultrasound (US) imaging system (AcousticX, Cyberdyne Inc., Ibaraki, Japan), referred to in this paper as LED-PAUS, is presented in Section 2.1 and then the modifications performed to use the same AcousticX data acquisition system for laser-illumination-based PA, referred to as laser-PAUS, without interrupting parameters of the imaging system, are presented in Section 2.2.

#### *2.1. LED Array-Based US/PA (LED-PAUS) Imaging System Description*

Conventionally, a high-power and bulky laser source is employed for most B-mode PA and US systems [28,29]. The commercial LED-PAUS system, as shown in Figure 1, consists of a host controller (Figure 1a), data acquisition hardware (Figure 1b) and an interactive graphical user interface shown on the display in Figure 1d. Here, a linear ultrasound probe is sandwiched between two LED arrays to capture interleaved B-mode PA and US images, as shown in Figure 1e. Each LED array consists of four rows of 36 LED elements (1 × 1 mm size). The LED arrays used in this study consist of 850 nm LED elements and each array provides an output energy of 200 μJ/pulse. The excitation pulse widths for each of these LED arrays can be controlled with the software in the range of 30 to 150 ns. For all the experiments in this study, a pulse width of 70 ns was used that offered optimum energy at 850 nm [56]. The optical illumination profile achieved with the two 850 nm LED arrays operated with pulse widths of 70 ns is shown in Figure 1g. The shape of the beam falling on skin is approximately a rectangle with an area of 9 cm<sup>2</sup> (5 by 1.8 cm), leading to an incident optical fluence of 0.044 mJ/cm2, considering 400 μJ total energy per pulse with the two LED arrays. The pulse repetition frequency (PRF) of these LEDs can also be controlled in the range of 1 to 4 KHz allowing multiple averaging options leading to different frame rates. To assess the effect of changing the frame rate over the PA image quality, several different combinations of PRF and frame averaging are used in this study, as discussed in the subsequent sections. The US probe used in this study is a 128-element linear US array with a pitch of 0.3 mm, center frequency of 7 MHz, elevational focus of 15 mm and a measured 6 dB bandwidth of 75%. The system provides PA and US acquisition sampling rates of 40 and 20 MHz, respectively.

#### *2.2. Experimental Setup for Comparing LED-PAUS and Laser-PAUS*

Figure 1 presents the overall experimental setup developed to compare the performance of LED-PAUS and laser-PAUS imaging. The LED-PAUS system, described in Section 2.1, was adapted for laser-PAUS imaging. In laser-PAUS, a portable optical parametric oscillator (OPO) laser source (Phocus Mobile, Opotek, Inc., Carlsbad, CA, USA), tunable in the range of 690–950 nm, shown in Figure 1c, provided the laser illumination. The laser has a fixed pulse width of 5–7 ns, a fixed PRF of 10 Hz and an output energy of 140 mJ per pulse at 750 nm. For this study, 850 nm wavelength laser illumination was used with the optical energy tuned down to 40 mJ/pulse. Light output from the laser was coupled to a 2 m long custom designed optical fiber bundle (Fiberoptic System Inc., Simi Valley, CA, USA). The fiber had a fused end with a diameter of 6.5 mm that entered into the tunable output port of the laser, as shown in Figure 1c. The distal end of this fiber was split into twenty smaller (1.45 mm inner diameter) fibers with numerical apertures of 0.55, each sharing equal optical energy. Ten out of these twenty fibers were inserted into each of the two custom designed 3-D printed fiber holders attached each side of the US probe, as shown in Figure 1f. This design allowed laser illumination similar to the case of LED arrays, in terms of the illumination angle and the overall geometry around the

US probe. Further, to achieve a uniform illumination profile on the tissue surface, two glass diffusers (N-BK7 Ground Glass Diffuser 1500 Grit, Thorlabs Inc., Newton, NJ, USA) were attached at the output end of the fiber holders, as shown in Figure 1f. The resulting laser-illumination profile at 850 nm is shown in Figure 1h. The shape of the beam falling on skin is approximately a rectangle with an area of 6 cm2 (5 by 1.2 cm), leading to an incident optical fluence of 6.66 mJ/cm2, considering 40 mJ total energy per pulse with the laser illumination. In our experimental setup, switching from LED arrays to a laser fiber setup and vice versa was convenient and did not disturb the US probe as well as the imaging subject. Figure 1i,j show the heads of LED-PAUS and laser-PAUS systems for acquiring respective B-mode PA and US images of human hand vasculature.

**Figure 1.** Description of the experimental setup designed for comparing light-emitting diode (LED)-based and highpower laser-based photoacoustic (PA) and ultrasound (US) imaging. The setup consists of the following key components: Commercial B-mode LED-PAUS system (AcousticX, Cyberdyne Inc., Ibaraki, Japan) with (**a**) a host controller PC and (**b**) data acquisition hardware. (**c**) A portable high-power laser (Phocus Mobile, Opotek Inc., Carlsbad, CA, USA) with its output coupled to the input end of an optical fiber bundle. The fiber bundle is split into twenty smaller fibers at the distal end. (**d**) Computer display: displays B-mode US (grayscale), PA (red scale), and coregistered US + PA (overlaid red PA on gray US). The interface also enables switching between LED and laser operation. (**e**) Arrangement of two 850 nm LED arrays around the US probe. (**f**) Arrangement of twenty laser fibers inserted into the two fiber holders around the US probe. Two glass diffusers attached at the fiber output ends to provide uniform laser illumination on the tissue surface. Optical illumination profile achieved with (**g**) two LED array sources and (**h**) laser source. (**i**,**j**) Pictures of a human wrist under imaging with the LED and laser arrangements, respectively.

#### **3. Validation Experiments and Results**

In this section, an extensive evaluation of the imaging performance of low-power LED-PAUS and high-power laser-PAUS systems is presented with the help of rigorous SNR and resolution studies on several tissue phantoms, in vivo imaging of the human wrist and in vivo 3-D vasculature mapping of the human forearm.

#### *3.1. Photoacoustic Imaging Comparison of LED Arrays and Laser Source Using a Scattering Phantom*

In this subsection, a controlled study evaluating the deep tissue PAI capabilities of lowpower LED arrays and a high-power laser source is presented. An acrylic tank with four holes was fabricated and a pencil lead with a diameter of 0.5 mm was inserted into each hole. The tank was filled with an intralipid solution to mimic an optical reduced scattering coefficient (*μ*- *<sup>s</sup>*) of 20 cm<sup>−</sup>1. Considering an approximate optical absorption coefficient (*μa*) of 0.05 to 0.1 cm<sup>−</sup>1, the effective attenuation coefficient (*μeff*) of the intralipid phantom was in the range of 1.73 to 2.45 cm−<sup>1</sup> (*μeff* = -3*μa*(*μ<sup>a</sup>* + *μ*- *<sup>s</sup>*)). Figure 2a shows the schematic of the experimental setup with four pencil lead targets diagonally arranged along the depth of imaging inside the intralipid solution. The measured depth of these four pencil leads from the surface of the US transducer were 15, 23, 28, and 34 mm, respectively.

The laser-illumination setup discussed in Section 2.2 was first employed to capture the PA images. The AcousticX data acquisition system software was synchronized to the laser acquisition mode. Two external triggers from the system were used for driving the laser. One trigger from the synch-1 port of the system went into the flash-lamp input port of the OPO laser and the other trigger from the synch-2 port fed the Q-Switch IN port of the laser, as shown in Figure 1b,c. After switching ON the 2-D B-mode RF data acquisition, the laser flash-lamp was first turned ON followed by (after 10s delay) the laser, from a PC connected to the laser. Once the laser was ON and synchronized with the data acquisition, PA data could be captured at a 10 Hz frame rate (limited by the PRF of the laser). Figure 2b shows a PA image captured with the laser illumination at 850 nm wavelength and 40 mJ/pulse optical energy resulting in <20 mJ/cm<sup>2</sup> optical fluence. As shown, all four pencil lead targets generated PA signals due to their higher optical absorptions compared to the intralipid medium.

To acquire the PA images with LED arrays, the laser arrangement was removed and the two 850 nm wavelength LED arrays were attached to the US transducer as shown in Figure 1e. The software of the AcousticX system was set to the LED acquisition mode. The PRF for the LED arrays was selected as 4 KHz. Figure 2c shows a PA image captured using the LED array setup with a frame averaging of 128. The achieved frame rate in this configuration was 30 Hz (128 averaging at 4 KHz PRF). Further, to study the effect of frame averaging over the LED-PA imaging performance, the averaging was increased from 128 to 256, 384, 640, 1280 and 2560 frames, leading to frame rates of 15, 10, 6, 3, and 1.5 Hz, respectively. The corresponding PA images are shown in Figure 2d–h.

In order to perform an effective comparison of the captured PA images with the laser and the LED array illuminations, the raw PA data were extracted and analyzed in a local computer using MATLAB software. First, frequency domain reconstruction [57] was performed to beamform the PA images from the raw data. All beamformed PA images were then log-compressed, maintaining the same 70 dB scale, as presented in Figure 2b–h, for effective comparison. Further, to quantitatively compare these PA images, the SNR study was performed over all four pencil lead targets. For calculating the SNR, peak PA signal at the target locations (over a circular region surrounding the targets) and the mean noise adjacent to each target (over a similar circular region at same depth as targets) were calculated over the linear beamformed PA images. The calculated values of peak signal, mean noise and SNR for the four targets over the PA images corresponding to the varying frame averaging of laser acquisition (1, 2, 4, 10, 128 frame averages leading to 10 Hz, 5 Hz, 2.5 Hz, 1 Hz, and 78 mHz frame rates when computed offline) and the LED array acquisition at varying frame rates are presented in Table 1. The calculations at 78 mHz with the laser system were specifically performed in order to compare the SNR of two systems at the same frame averaging—i.e., 128 frames (LED 30 Hz). As shown, the peak PA signal for the laser acquisition is about two log orders higher than that for the LED arrays. However, the mean noise in the case of laser illumination is even higher (up to three log orders), leading to a lower SNR, especially when compared at shallow imaging depths.

Plots in Figure 2i–l present the SNR trends observed with varying frame rates of LED array acquisitions for the four targets. The corresponding value of SNR for the laser acquisition at 10 Hz is also marked in each of these plots. For the targets lower than 30 mm depth, the LED-based PAI at 10 Hz frame rate continued to show a higher SNR. Target-4 at 34 mm depth was detected at a higher SNR with the laser illumination as compared to the LED arrays at the same frame rate. However, increasing the PA frame averaging further for the LED arrays acquisition led to a lower frame rate (<10 Hz), and helped in boosting the SNR value for higher depth targets. Further, when maintaining same frame averaging (128 frames) for the laser (78 mHz frame rate) and the LED arrays (30 Hz frame rate), significantly high SNRs were observed for all four targets with the laser, sacrificing the real-time imaging.

**Figure 2.** Performance evaluation of LED array-based and laser-illumination-based photoacoustic (PA) imaging in intralipid scattering phantom. (**a**) Shows the schematic of a scattering phantom with four 0.5 mm diameter pencil leads placed at 15, 23, 28 and 34 mm depth from the ultrasound (US) transducer inside intralipid medium mimicking an optical reduced scattering coefficient of 20 cm<sup>−</sup>1. (**b**) Shows the PA imaging results with 850 nm laser illumination at 10 Hz and 40 mJ optical energy with <20 mJ/cm<sup>2</sup> optical fluence on the phantom surface. (**c**–**h**) Show the PA imaging results with LED array-based illumination at an 850 nm wavelength, a total of 400 μJ output energy from the two LED arrays and frame rates of 30, 15, 10, 6, 3 and 1.5 Hz, respectively. (**i**–**l**) Show the plots of signal-to-noise ratio (SNR) with respect to the acquisition frame rates comparing LED array and laser-illumination-based PA imaging performance for the pencil lead targets at four depths.


**Table 1.** Peak PA signal, average background noise and signal-to-noise ratio (SNR) values for four pencil lead targets located 15, 23, 28 and 34 mm deep inside scattering phantom imaged with LED array and laser illuminations at varying frame rates.

> In order to validate the optical properties of our phantom, we calculated the *μeff* value from the experimental peak PA signal values for LED 10 Hz data. Using the Beer Lambert's principle, i.e., *I*(*z*) = *Ioe* −*μeff z* , where *I(z)* is the PA intensity at depth (z) cm and *Io* is the intensity at zero depth, and taking the ratio of two equations at two different depths (z1 at 1.5 cm and z2 at 3.4 cm) will cancel out the *Io* and lead to the *μeff* 2.034 cm<sup>−</sup>1. This closely matches to the previously reported values in the range of 1.734 to 2.456 cm<sup>−</sup>1, mentioned above in the phantom description.

#### *3.2. Photoacoustic Imaging Comparision of LED Arrays and Laser Illuminations over Chicken Tissue Phantom*

To compare the deep tissue PAI capabilities of the low-power LED arrays and the highpower laser illuminations, a multilayer chicken tissue phantom was designed. Figure 3a shows the schematic of the chicken tissue phantom with five layers of chicken breast tissue stacked inside a water tank, with an estimated optical absorption and reduced scattering coefficients of 0.1 to 0.2 cm−<sup>1</sup> and 1.0 to 5 cm−1, respectively, with an 850 nm wavelength [58], leading to an effective attenuation coefficient (*μeff*) in the range of 0.575 to 1.766 cm−<sup>1</sup> (*μeff* = -3*μa*(*μ<sup>a</sup>* + *μ*- *<sup>s</sup>*)). Four pencil leads with diameters of 0.5 mm were placed in between the chicken tissue layers as shown. The measured depths of these pencil lead targets from the top layer of the chicken tissue were 11, 18, 24, and 31 mm, respectively.

With the above-described phantom, a laser-illumination setup was used to acquire the US and PA images at 10 Hz, maintaining the same 40 mJ output energy (<20 mJ/cm2 optical fluence on the phantom surface) and laser with an 850 nm wavelength. The captured raw data from the AcousticX software were reconstructed in MATLAB to further perform the quantitative comparison. Figure 3b shows the beamformed B-mode US image of the chicken tissue phantom, at a log scale of 70 dB, clearly highlighting the chicken tissue structure. The US image also shows the distance of the four pencil lead targets from the top layer of chicken tissue. Similarly, the PA raw data captured from the software were reconstructed, log-compressed and was overlaid on the US image to generate a coregistered US + PA image of the phantom. Figure 3c shows the coregistered US and PA image at a 60 dB scale in order to keep the noise floor at the threshold. The beamformed PA image with 70 dB log scale is shown in Figure 3d.

Without disturbing the phantom, the PA images were subsequently acquired with the LED arrays setup. The laser arrangement was removed and the two 850 nm LED arrays were attached to the US probe. With the 4 KHz PRF of the LED arrays, the PA images were captured at varying frame averaging settings, similar to the settings used for Section 3.1, leading to the frame rates of 30, 15, 10, 6, 3 and 1.5 Hz. The captured raw data from the AcousticX software were again extracted in the MATLAB software and reconstructed to generate B-mode PA images for comparing with the laser PA images. All PA images were compressed at the 70 dB log scale for effective comparison. Four representative PA images at 30, 15, 10, and 6 Hz, respectively, are shown in Figure 3e–h.

**Figure 3.** Performance evaluation of LED array-based and laser-based photoacoustic (PA) imaging in chicken tissue phantom. (**a**) Shows a schematic of chicken tissue phantom with five layers of chicken breast tissue stacked inside water tank. The positions of four 0.5 mm diameter pencil lead targets placed in between the chicken tissue layers are also shown. (**b**–**d**) Show the B-mode ultrasound (US), coregistered US + PA and PA image, respectively, captured with laser-based illumination at 850 nm wavelength and 10 Hz high pulse repetition frequency (PRF). (**e**–**h**) Show the PA imaging results obtained by imaging the chicken tissue phantom using LED array-based illumination, at 850 nm illumination and frame rates of 30, 15, 10, and 6 Hz, respectively. (**i**) Shows the plot of signal-to-noise ratio (SNR) with respect to the frame rate comparing the LED array and laser-illumination-based PA imaging performance for the deepest pencil lead target (target-4).

Further, for a quantitative comparison of high-power laser versus low power LED array-based acquisitions over the chicken tissue phantom, the SNR study was performed over the linear beamformed images in MATLAB. For this experiment, we studied the SNR of the shallowest target (target-1 located 11 mm) and the deepest target (target-4 located 31 mm) deep inside chicken breast tissue. For calculating the SNR, the peak signal at target locations and the mean noise adjacent to these targets were calculated. The values of peak PA signal, mean adjacent noise and the SNR for laser-based and LED array-based acquisitions at varying frame rates are listed in Table 2. The peak signal as well as mean noise for laser acquisition at 10 Hz is about two to three log orders of magnitude higher compared to the LED arrays. However, the SNR for LEDs is still comparable to the lasers. Figure 3i also demonstrates the trend of SNR for target-4, comparing the laser and the LED array illuminations at varying frame rates. For a 31 mm deep target, LED arrays provide

close to 37.51 dB SNRs, whereas laser illumination provides about 43.75 dB SNR at a 10 Hz frame rate. When increasing the frame averaging, hence a reduction in the frame rate, LED arrays show significant improvement of SNR for the same target, up to 54.47 dB at 1.5 Hz, which is higher than the SNR of laser illumination at 10 Hz. These results demonstrate the capabilities of LED arrays to image deeper inside realistic tissue medium and motivated us to further study how they compare with high-power laser sources for imaging in vivo human vasculature.

**Table 2.** Peak PA signal, average background noise and signal-to-noise ratio (SNR) values for pencil lead targets located 11 and 31 mm deep inside chicken breast tissue imaged with LED array and laser illuminations at varying frame rates.


Similar to Section 3.1, we validated the optical properties of this chicken tissue phantom by calculating the *μeff* value from the experimental peak PA signal values for LED 10 Hz data. The experimentally calculated *μeff* (2.067 cm−1) is slightly higher than the previously calculated range mentioned above in the phantom description. The small discrepancy could be due to chicken tissue heterogeneity—prolonged imaging of chicken tissue in water medium that increases the optical attenuation.

#### *3.3. Photoacoustic Imaging Comparison of LED Arrays and Laser Sources: Resolution Study*

In this subsection, the spatial resolutions of the two optical illumination setups, the LED-PAUS and the laser-PAUS, are characterized. A 30 μm carbon fiber was placed in a bath containing water mixed with intralipid to obtain a scattering medium of 3 cm<sup>−</sup>1. First, the two 850 nm LED arrays were attached to the US probe and a B-mode PA image of the phantom was acquired. The raw data captured were extracted in the MATLAB software and reconstructed to generate a B-mode image. The log-compressed B-mode PA image at a 30 dB scale is shown in Figure 4a. Figure 4b presents a sample zoomed time trace of an A-line across the target region for the PA data acquired with LED array illumination. To calculate the spatial resolution for this carbon fiber target, the line-spread functions of the PA amplitudes are plotted in the lateral and axial directions, respectively, as shown in Figure 4c,d. The obtained lateral and axial resolutions with a full-width-half-maximum (FWHM) approach are 350 and 210 μm, respectively.

To acquire the PA data with laser-illumination, the LED arrays were removed from the US probe and the laser fiber setup was attached without disturbing the phantom. The captured PA data were extracted and beamformed. Figure 4e shows the B-mode PA image, at a 30 dB scale, obtained for the same carbon fiber phantom using laser-illumination at 850 nm with 40 mJ output energy and a 10 Hz frame rate. A sample zoomed time trace of an A-line across the target region for the PA data acquired with the laser-illumination is shown in Figure 4f. The line-spread functions of the PA amplitudes in the lateral and axial directions with the laser illumination are shown in Figure 4g,h. The obtained lateral and axial resolutions using an FWHM approach for the laser illumination are 355 and 203 μm, respectively.

**Figure 4.** Resolution study for LED array-based and laser-illumination-based photoacoustic (PA) imaging. (**a**) B-mode PA image obtained for a 30 μm carbon fiber placed in an intralipid-based phantom using LED illumination at 850 nm with a frame rate of 10 Hz. (**b**) Shows zoomed time trace of an A-line across the target region for the PA data acquired with LED illumination. (**c**,**d**) Show the line-spread functions of the PA amplitudes for the carbon fiber target plotted in the lateral and axial directions, respectively, when imaged with LED setup. The obtained lateral and axial resolutions with the full-width-half-maximum (FWHM) approach are 350 and 210 μm, respectively. (**e**) B-mode PA image obtained for same carbon fiber phantom using laser illumination at 850 nm and 10 Hz frame rate. (**f**) Shows the time trace for the PA data acquired with laser illumination. (**g**,**h**) Show the line-spread functions of the PA amplitudes in lateral and axial directions with the laser illumination. The obtained lateral and axial resolutions using the FWHM approach are 355 and 203 μm, respectively.

#### *3.4. In Vivo Photoacoustic Imaging Comparison of LED Arrays and Laser Illumination over In Vivo Human Wrist*

In this subsection, the photoacoustic vascular imaging capabilities of the LED array and the laser illumination were compared by in vivo imaging of a healthy human volunteer's wrist vasculature. The volunteer was a healthy 25-year-old European male, and the experiment was conducted by following the internal imaging protocol of CYBERDYNE, INC (Rotterdam, The Netherlands) for healthy-volunteer imaging studies. For this study, the volunteer's right hand was positioned inside a large water tank, as shown in Figure 5a. The probe was positioned such that one of the major blood vessels, which supplies blood to the forearm and hand, is within the field-of-view (FOV).

The hand was first imaged with the laser-PAUS setup by attaching the laser fiber holders to the US probe, as shown in Figure 5a. The laser was operated at 850 nm wavelength, 10 Hz PRF, and delivered output optical energy of 40 mJ. This allowed ANSI safety limits of <20 mJ/cm2 optical fluence on the hand surface [59]. During the real-time data acquisition, the probe was aligned such that the major blood vessel could be seen running parallel to the skin surface in the PA images. The captured US and PA raw data using the AcousticX software were later extracted in the MATLAB software and were reconstructed to generate the beamformed images. The beamformed, log-compressed B-mode US, PA and the coregistered US + PA images are shown in Figure 5b–d. The US image showed the anatomical features along the depth of human wrist, whereas the PA image highlighted the major blood vasculature. There was fairly strong correspondence between the locations of the blood vessel in the PA image and the appearance of anechoic regions in the US image. Based on the anatomy of the vasculature in human wrist, the PA signals ~5 mm below the skin surface may have corresponded to the radial artery that travels across the front of the elbow, deep under the muscle until it comes to the wrist where it comes close to the skin surface. This is also marked with a white arrow in the PA image in Figure 5c.

To compare these laser-illumination results of the human wrist with the LED arrays, the laser-fiber attachments were gently removed without disturbing the location of the US probe. The two 850 nm LED arrays were then attached to the US probe, as shown in Figure 5e. With a PRF of 4 KHz and frame averaging of 384, leading to a frame rate of 10 Hz, the US and PA frames were captured using the LED array setup. The US and PA raw data were then reconstructed in MATLAB. Figure 5f–h show the beamformed log-compressed B-mode US, PA and coregistered US + PA images for the human wrist. As in the case of laser illumination, the LED array-based PA images also imaged the same vasculature below the skin surface. The radial artery present ~5 mm below the skin was clearly visible with the LED array-based acquisition as well.

**Figure 5.** In vivo comparison of LED array-based and laser-based PA vasculature imaging over the right-hand wrist of a healthy 25-year-old male human volunteer. (**a**) Shows the experimental setup with the right-hand wrist placed inside a big water bath for the laser-based PA imaging. (**b**–**d**) Show the obtained US, PA and coregistered US + PA images for the setup shown in (**a**). (**e**) Shows the setup with LED arrays. (**f**–**h**) Show the obtained US, PA, and coregistered US + PA images for the setup shown in (**e**).

To further compare the two setups quantitatively, an SNR comparison study was performed for the radial artery, as marked with white arrows in Figure 5c,g. To calculate the SNR, the peak PA signal at the artery and the mean noise adjacent to the artery region was calculated over the linear beamformed PA images resulting from the laser-illumination and the LED array-based acquisitions. Table 3 presents the values of peak PA signal, mean noise and the SNR. The SNR values are also marked in the Figure 5c,g. Both the peak signal and mean noise with the laser are up to three log orders of magnitude higher compared to the LEDs. However, the SNR value for the laser-illumination-based PA image was about 6 dB lower than the SNR with the LED array acquisition. This follows the trend observed for the controlled tissue phantom studies discussed in Section 3.1, where the shallow targets (<30 mm) inside an intralipid medium were detected with higher SNRs using LED arrays compared to the laser illumination, maintaining the same frame rate.


**Table 3.** Peak PA signal, average background noise and signal-to-noise ratio (SNR) values for blood vessel target located ~5 mm below the skin surface of right-hand wrist of a healthy 25-year-old male human volunteer imaged with LED array and laser illumination at 10 Hz frame rate.

*3.5. In Vivo Photoacoustic Imaging Comparison of LED Arrays and Laser Illumination: In Vivo Human Forearm*

This subsection presents in vivo 3-D mapping of the vasculature inside a human volunteer's forearm, using the LED array-based and the laser-based illuminations. For this study, the volunteer was a healthy 25-year-old European male, and the experiment was conducted by following the internal imaging protocol of CYBERDYNE, INC (Rotterdam, The Netherlands) for healthy-volunteer imaging experiments. The forearm of the volunteer was submerged in water. The US probe was fixed on a linear translation stage, translating in the Y-direction, as shown in Figure 6a.

For the laser acquisition, the laser-illumination setup was attached to the US probe. After switching ON the laser software, the RF data acquisition on the AcousticX software were first turned ON. Note that the 3-D data acquisition feature of the AcousticX software does not work for the laser mode. Therefore, to capture the data corresponding to the Y-direction scan, a manual stage motion feature was used while acquiring the data in the 2-D RF acquisition mode. The linear stage was translated for 60 mm in the Y-direction while capturing the RF data. The laser was tuned to 850 nm at 40 mJ output energy (<20 mJ/cm<sup>2</sup> optical fluence on the hand surface) and 10 Hz PRF. The captured PA raw data were later extracted and reconstructed in MATLAB. Initial PA frames captured prior to the stage motion were discarded for the 3-D volumetric reconstruction. Figure 6b presents the maximum-intensity-projection (MIP) of the 3-D PA volume for full depth, highlighting the major vasculature present in the human forearm. To better visualize the vasculature deeper than ~5 mm, a deep tissue MIP of the PA volume is also shown in Figure 6c. Considering the initial 15 mm stand-off of the US probe from the human skin surface, the deep tissue MIP fell within Z = 20 to 40 mm, where Z is the depth dimension. Since deeper blood vessels show weak PA intensities, the MIP image in Figure 6c was scaled to 30 dB as opposed to the full depth MIP in Figure 6b, which was scaled to 50 dB.

After the laser acquisition, the stage was manually moved back to the home position and the laser setup was detached from the US probe without disturbing the position of the forearm with respect to the probe. To acquire the LED array-based PA 3-D scan data, two 850 nm LED arrays were attached to the US probe. In this case, the AcousticX software directly allows 3-D PA data acquisition using the automatic 3-D scan feature. The same Ydirection translation stage was automatically scanned for 60 mm and the raw data captured were extracted in MATLAB. The beamformed PA frames were stitched in the Y-dimension to reconstruct a 3-D PA volume. Figure 6d,e present the full depth MIP and the deep tissue MIP of the 3-D PA volume, respectively.

Qualitatively, Figure 6b,e look very similar, indicating that both laser and LED arraybased illumination are able to map the major vasculature present inside human forearm. Minor differences in the SNR for deep vessels can be noticed when comparing the deep tissue MIPs presented in Figure 6c,f. For example, a deeper blood vessel running parallel to the *Y*-axis at around X = 20 mm (in Figure 6c,f) was detected with a better SNR with the laser source as compared to the LED arrays. Similarly, the deep tissue MIP from laser acquisition show other minor PA signals (at coordinates X,Y = 30 mm, 50 mm), that are relatively weaker in the deep tissue MIP from the LED acquisitions, emphasizing the need for a high-power laser when imaging very deep (>2 cm) vasculature.

**Figure 6.** In vivo comparison of LED array-based and laser-based PA vasculature imaging over forearm of a healthy 25-year-old male human volunteer. (**a**) Shows the experimental setup with the right forearm placed inside big water bath for the laser-based PA imaging. (**b**,**c**) Show the obtained full depth and deep tissue maximum-intensity-projections (MIPs) of the reconstructed 3-D PA volume for the setup shown in (**a**). (**d**) Shows the setup with LED arrays. (**e**,**f**) Show the obtained full depth and deep tissue MIPs of the reconstructed 3-D PA volume for the setup shown in (**d**).

#### **4. Discussion**

To successfully translate PAI to the POC applications and to resource-limited settings, the cost and the overall size of the PAI systems need to be significantly cut down. LED arrays are one of the highly explored optical sources in the PA literature that are portable and significantly lower in cost and footprint. This study was focused on comparing the capabilities of these LED arrays (approximately USD 15K, including driver electronics) with a state-of-the-art high-power OPO laser (approximately USD 100K) for PAI. A commercial LED-PAUS system, AcousticX, was first adapted to perform sequential acquisitions of LED-PAUS and laser-PAUS. This ensured that the PA and US signals were detected by the same ultrasound transducer array and processed by the same data acquisition system. To further draw a fair comparison, the experimental setup shown in Figure 1 allowed for (1) similar optical illumination on the tissue surface in terms of the geometry, angle and aperture of illumination, and (2) a convenient, uninterrupted sequential acquisition of PA images with the two sources. To achieve uniform laser illumination on the tissue surface, two glass diffusers were attached to the output end of the laser fiber holders. For all the experimental studies presented in this work, the output optical energy for the laser and the LED arrays were maintained at 40 mJ and 400 μJ/pulse, respectively, with an 850 nm wavelength of illumination. A constant PRF of 10 Hz for the laser and 4 KHz for the LED arrays were used for all studies. With no frame averaging, the obtained PA frame rate for laser was 10 Hz. For the LED arrays, the PA frame averaging was varied from 128 to 256, 384, 640, 1280, and 2560, achieving the frame rates of 30, 15, 10, 6, 3, and 1.5 Hz, respectively.

The first comparison study presented in Section 3.1 consisted of controlled experiments on an intralipid phantom shown in Figure 2. The four 0.5 mm pencil lead targets arranged diagonally along the depth of the scattering phantom were imaged using both laser and LED arrays. Detailed analysis of the SNR presented in Table 1 and plotted in Figure 2i–l led to the following conclusions on the quantitative performance of the two sources. (1) At similar frame rates (10 Hz), LED-PAUS imaging showed higher SNR compared to laser-PAUS imaging for targets up to an imaging depth of 30 mm. (2) Laser-PAUS imaging provided a better SNR for the deeper targets (>30 mm) as compared to the LED-PAUS at 10 Hz. (3) With increased frame averaging and hence a reduced overall frame rate (<10 Hz),

the deep tissue (>30 mm) performance of LED arrays can be improved. For example, the target-4 located 34 mm deep was imaged with a 48.24 dB SNR using laser illumination and with 44.22 dB SNR using LED arrays, both at 10 Hz. However, with increase in PA averaging (lower frame rates), LED arrays were able to provide an improved SNR of up to 50.92, 55.79, and 63.55 dB, when imaged at 6, 3, and 1.5 Hz frame rates, respectively. Further, comparison of the magnitudes of peak PA signals from the target and mean noise surrounding the target for the two optical illuminations led to the following observations. While higher optical energy from the laser source generates a PA signal two to three log orders of magnitude higher compared to the lower-power LED arrays, the average noise floor from the background regions was also observed to be up to three to four log orders higher for the laser sources compared to LED arrays. In conclusion, the low-power LED arrays were effective due to lower noise floor that comes from the higher frame averaging possible with high PRF (4 KHz) of LED arrays.

Our next comparison study, presented in Section 3.2, involved a multilayer chicken tissue phantom with four 0.5 mm pencil lead targets embedded in between the five layers of chicken breast tissue. Figure 3 presented the qualitative comparison of laser-PAUS and LED-PAUS, whereas Table 2 highlighted the quantitative comparison of PAI performance for the deepest pencil lead target—i.e., target-4 at 31 mm. This study results further confirmed our observations with the intralipid scattering phantom presented in Section 3.1. For example: (1) both peak PA signal and mean noise from the laser illumination were up to three log orders of magnitude higher compared to the LED array illumination. (2) For the target-4 at 31 mm, the laser showed a ~6 dB higher SNR compared to the LED arrays at a 10 Hz PA frame rate. (3) With the increase in the average, and hence reduced, frame rate (<10 Hz), LED arrays show higher SNR values than laser illumination, obtaining SNRs of 54.47 and 47.76 dB at 1.5 and 3 Hz, respectively, for the same target-4 as compared to an SNR of 43.75 dB with the laser.

The study presented in Section 3.3 compared the lateral and axial resolutions of the two setups, LED-PAUS and laser-PAUS, by imaging carbon fiber with a 30 μm diameter inside an intralipid-based phantom. Although the pulse widths for the laser (5–7 ns) and the LED arrays (70 ns) in this study were not similar, the spatial resolution were shown to be in the same ranges—i.e., ~350 μm lateral and ~205 μm axial. This confirmed the fact that the spatial resolution here is mainly limited by the acoustic detection.

In the final study, in Section 3.4 we presented an in vivo LED-PAUS and laser-PAUS imaging of the wrist of a healthy human volunteer. Figure 5 presents the qualitative and quantitative comparison results for a radial artery seen ~5 mm below the wrist skin surface. Analysis of the peak PA signal, mean noise and SNR for this radial artery is presented in Table 3. As observed with the studies presented in previous sections, for a shallow depth target, the LED-PAUS showed better SNR than the laser-PAUS at the same 10 Hz frame rates. The ~5 mm deep radial artery was imaged with a 36.37 dB SNR with the laser compared to a 42.49 dB SNR with LED arrays. This study substantiates that the LED-PAUS imaging is an attractive choice for several preclinical and clinical applications.

Further, in Section 3.5, to test the capabilities of deep tissue in vivo vascular imaging, we scanned the right-hand forearm of a healthy human volunteer. The 3-D vasculature map, presented in Figure 6, demonstrated that the LED arrays could image all the vessels that could be seen with a laser source, especially at shallow depths of up to 1.5 cm. For deeper (>30 mm) vessels, laser sources provided better signals. However, the amount of noise observed with the laser at those depths was also significantly high and thus limited the imaging performance when viewing deeper vessels. With an increased frame averaging with the LED arrays, the SNR of deeper vessels can potentially be improved.

The conclusive observations made by studying the performance of the two optical sources bring us to the following remarks. (1) LED arrays exhibit a strong potential for translating PAI systems to the resource-limited settings. (2) Higher frame averaging enhances the LED-PAUS imaging performance, especially for deeper targets (>30 mm), making it suitable for deep tissue imaging; however, this sacrifices the frame rate. To further improve the performance of LED-PAUS imaging, the state-of-the-art machine learning approaches can be employed that can help in (1) further boosting the frame rates by avoiding the need for higher frame averaging [60,61], and (2) improving the SNR for deep tissue targets [61,62].

#### **5. Conclusions**

Photoacoustic imaging capabilities of low-cost and low-power LED arrays were compared head-to-head with a high-power laser source using both tissue-mimicking phantoms and in vivo human subjects. The experimental observations on different tissue mimicking phantom studies demonstrated (1) high PRFs of LED arrays can be leveraged for averaging PA frames to achieve high SNR up to an imaging depth of 30 mm with 10 Hz frame rate, (2) high-power laser sources show higher SNRs for deeper targets (>30 mm) compared to the LED arrays at 10 Hz and (3) with increased frame averaging (<10 Hz), the SNR of a target at 34 mm depth with LED-PAUS closely matched that of laser-PAUS. The in vivo human hand vasculature imaging studies presented in this work also demonstrated similar observations. (1) The ~5 mm deep radial artery was imaged with a higher SNR using LED arrays in comparison to the laser source, and (2) for deeper vessels in the subject's forearm, the laser at the 10 Hz frame rate had a slightly higher SNR compared to the LED arrays at 10 Hz. In summary, due to the low power of LED arrays, a higher frame averaging is required to image deep tissue targets. LED-PAUS holds strong potential in point-of-care diagnosis of vascular diseases.

**Author Contributions:** Conceptualization: S.A., M.K.A.S. and S.-R.K.; Methodology: S.A., and K.J.; Software: S.A.; Hardware design: S.A.; Validation: S.A., M.K.A.S. and K.J.; Formal analysis: S.A. and M.K.A.S.; Investigation: S.-R.K.; Resources: S.-R.K.; Data curation: S.A. and M.K.A.S.; Writing– original draft preparation: S.A. and M.K.A.S.; Writing–review and editing: S.A., M.K.A.S., K.J., D.C.H., C.R.P., and S.-R.K.; Visualization: S.A., and K.J.; Supervision: S.-R.K.; Project administration: S.-R.K.; Funding acquisition, S.-R.K. All authors have read and agreed to the published version of the manuscript.

**Funding:** This project was funded by the NIH-NIBIB R00EB017729-04 (S.-R.K.), R21EB030370-01 (S.-R.K.) and support of Penn State Cancer Institute (S.-R.K.).

**Institutional Review Board Statement:** The human volunteer studies were conducted by following the internal imaging protocol of CYBERDYNE, INC (Rotterdam, The Netherlands) for healthyvolunteer imaging experiments, adhering to the guidelines of the Declaration of Helsinki.

**Informed Consent Statement:** Informed consent was obtained from all the subjects involved in this study.

**Data Availability Statement:** Data available on request from the authors.

**Acknowledgments:** We gratefully acknowledge the support from Gary Meyers and Eugene Gerber, for their help with the 3-D printing work and the machining of experimental setup parts.

**Conflicts of Interest:** M.K.A.S. is employed by CYBERDYNE, INC. The author(s) declare no potential conflicts of interest with respect to the research, authorship and/or publication of this article.

#### **Abbreviations**


#### **References**


## *Article* **Remote Photoacoustic Sensing Using Single Speckle Analysis by an Ultra-Fast Four Quadrant Photo-Detector**

**Benjamin Lengenfelder 1,2,\*, Martin Hohmann 1,2, Moritz Späth 1,2 , Daniel Scherbaum 1, Manuel Weiß 3, Stefan J. Rupitsch 4, Michael Schmidt 1,2, Zeev Zalevsky 2,5 and Florian Klämpfl 1,2**


**Abstract:** The need for tissue contact makes photoacoustic imaging not applicable for special medical applications like wound imaging, endoscopy, or laser surgery. An easy, stable, and contactfree sensing technique might thus help to broaden the applications of the medical imaging modality. In this work, it is demonstrated for the first time that remote photoacoustic sensing by speckle analysis can be performed in the MHz sampling range by tracking a single speckle using a four quadrant photo-detector. A single speckle, which is created by self-interference of surface back-reflection, is temporally analyzed using this photo-detector. Phantoms and skin samples are measured in transmission and reflection mode. The potential for miniaturization for endoscopic application is demonstrated by fiber bundle measurements. In addition, sensing parameters are discussed. Photoacoustic sensing in the MHz sampling range by single speckle analysis with the four quadrant detector is successfully demonstrated. Furthermore, the endoscopic applicability is proven, and the sensing parameters are convenient for photoacoustic sensing. It can be concluded that a single speckle contains all the relevant information for remote photoacoustic signal detection. Single speckle sensing is therefore an easy, robust, contact-free photoacoustic detection technique and holds the potential for economical, ultra-fast photoacoustic sensing. The new detection technique might thus help to broaden the field of photoacoustic imaging applications in the future.

**Keywords:** photoacoustic; remote sensing; endoscopy; speckle

#### **1. Introduction**

Photoacoustic imaging (PA) is a new, rising imaging technique since it combines a high penetration depth with a good image contrast [1]. As PA is based on the absorption of a short light pulse and the subsequent acoustic signal generation, the absorption characteristics define the imaging contrast. As the absorption coefficients for different tissue absorbers differ greatly, it is possible to obtain a high image contrast. Hence, PA is especially attractive for displaying blood vessels due to the high hemoglobin absorption compared to other tissue constituents in the visible and near-infrared range [2]. This high contrast is combined with a high penetration depth of several millimeters since the acoustic scattering is up to a factor of 1000 less than the optical scattering [3]. Thus, higher imaging depths than for purely optical methods can be achieved with ease.

**Citation:** Lengenfelder, B.; Hohmann, M.; Späth, M.; Scherbaum, D.; Weiß, M.; Rupitsch, S.J.; Schmidt, M.; Zalevsky, Z.; Klämpfl, F. Remote Photoacoustic Sensing Using Single Speckle Analysis by an Ultra-Fast Four Quadrant Photo-Detector. *Sensors* **2021**, *21*, 2109. https:// doi.org/10.3390/s21062109

Academic Editors: Mithun Kuniyil Ajith Singh and Wenfeng Xia

Received: 22 January 2021 Accepted: 11 March 2021 Published: 17 March 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

Since piezo element transducers, which are contact based, are the state-of-the-art detection technique, PA requires tissue contact. As a consequence, PA may not be suitable for special medical applications, such as wound imaging or laser surgery. Furthermore, transducer miniaturization is challenging, and small elements suffer from low sensitivity, which limits their usage for photoacoustic endoscopy. Therefore, a remote and stable detection system would be beneficial.

Air-coupled transducers could overcome this issue. However, they suffer from poor sensitivity and are still too big for minimally-invasive medical applications [4]. For the noncontact beam deflection technique, the deflection of a probe beam that is positioned above the tissue is monitored by a position sensitive detector [5,6]. A small fraction of the photoacoustic signal is transmitted to the surrounding air and deflects the probe beam. Since the probe beam needs to be scanned above the whole tissue surface, this modality is unsuitable for surgical usage. In addition, the detection bandwidth is limited due to the probe beam size. Interferometric detection methods offer non-contact sensing and a higher detection bandwidth. Here, the surface displacement after photoacoustic signal generation is monitored for the initial pressure reconstruction inside the object [7–10]. Nevertheless, interferometric systems are expensive and require a complicated setup that is noise sensitive [11]. In addition, interferometric setups only monitor the surface displacement correctly, if the surface is properly tilted, which cannot be guaranteed for clinical application. There are also non-interferometric approaches for remote acoustic signal detection. Clark et al. detected surface acoustic waves by tracking the movement of multiple speckles using simple diode detection systems [12,13]. For this approach, however, the speckles are imaged in the near-field, and thus, Fresnel diffraction applies. In this case, the movement of the speckles is dependent on three types of movement, which cannot be separated and occur simultaneously: transverse, axial, and tilt [14]. Due to this disadvantage, the acoustic sensing approach of Clark et al. may not be suitable for photoacoustic imaging. Hajireza et al. sensed the photoacoustic signal directly at its origin by a non-interferometric system [15]. This system monitors the reflection of a probe beam that is sensitive to the elasto-optic index modulation induced by the photoacoustic initial pressure transients. This approach allows non-interferometric, non-contact photoacoustic sensing. However, it provides only penetration depths in the mm-range due to the high optical attenuation for the probe beam. In addition, it is only applicable for photoacoustic microscopy and not suited for photoacoustic tomography since it does not provide temporally resolved acquisition of the acoustic signal and thus requires depth scanning for image acquisition.

Remote speckle analysis is an easy, robust, and non-interferometric vibration sensing technique whose potential for photoacoustic tomography we already demonstrated in previous publications [16–18]. By tracking the movement of multiple speckles, it was possible to remotely reconstruct the photoacoustic signal. However, in these proof-ofconcept studies, the acoustic sensing bandwidth and thus resolution were limited to the frame rate of the expensive, high-resolution camera used at approximately 800 kHz.

In the present publication, it is demonstrated for the first time that remote photoacoustic sensing with a sampling rate of 8 MHz can be performed by tracking the movement of a single speckle with only four diodes. The speckle tracking by only four diodes proves the applicability of an economical position sensitive diode for speckle analysis. Polymer phantoms and skin tissue samples are measured in transmission mode and reflection mode. In addition, the capability of easy miniaturization and endoscopic usage of the single speckle analysis is shown by fiber bundle measurements. The new technique is an essential step for the implementation of a remote photoacoustic imaging system using speckle analysis. Therefore, this work might help to broaden the applications of PA in special applications like wound imaging, endoscopy, or guiding laser surgeries.

#### **2. Materials and Methods**

*2.1. Speckle Sensing by Multiple Speckles' Tracking*

The speckle sensing technique is based on the time-resolved detection of the position of a speckle pattern. It is possible to extract the tilt change of a laser illuminated surface by tracking the speckle pattern movement [14]. The left side of Figure 1 shows a tilting object surface, and the right side illustrates the speckle sensing technique.

**Figure 1.** Speckle sensing theory: An object tilt results in a speckle pattern movement, which can be reconstructed with an imaging system. Figure after [19].

If the observation plane distance *Z* for the generated objective speckle pattern fulfills the far-field approximation (*Z* > (*D*<sup>2</sup> *ill*)/(4*λ*), illuminated diameter *Dill*), the objective speckle pattern movement (*δo*) is only dependent on the surface tilt change Δ*α* and *Z*. By imaging the objective speckle pattern using an imaging system with the magnification *M*, a subjective speckle pattern is created on the imaging sensor whose lateral movement (*δs*) is also linearly proportional to Δ*α* assuming small angle approximation. Equation (1) explains the relation between the speckle movements, *M* and *Z* [19].

$$
\delta\_s = \delta\_o \cdot M = \tan(\Delta a) \cdot Z \cdot M \tag{1}
$$

By acquiring a speckle pattern video sequence with the imaging system, it is thus possible to measure the surface tilt. Correlation analysis of the image frames allows the reconstruction of the speckle pattern movement along the sensor axes x and y, which are representative of the surface tilts along these two axes' directions [19]. In this article, it is shown for the first time that speckle sensing is feasible by tracking the movement of a single speckle in contrast to the described correlation analysis of a multiple speckle image. Therefore, it is proven that a single speckle contains all necessary information for photoacoustic signal detection.

#### *2.2. Imaging Systems and Setups*

Two imaging systems are established for this work and described here in detail: the free-space setup, which is capable of far distance PA sensing, and the fiber based setup, which is suitable for endoscopy.

#### 2.2.1. Free-Space Single Speckle Sensing

Figure 2 shows the established diode based, free-space imaging system. The imaging system consists of a microscope objective (10×), a beamsplitter and a lens (*f* = 100 mm) used for image magnification on a camera (DCC1545M, Thorlabs, Newton, NJ, USA), and an avalanche-photodiode sensor (APS; APDcam, Fusion Instruments, Budapest, Hungary). A bandpass filter for the speckle wavelength is placed after the objective in order to block the photoacoustic excitation and room light. The magnification of the system is calculated at 5 with a microscope test target (*M* = 5). The camera is used as a reference for image calibration and alignment. The calibration is done with a multi-mode fiber (AFS105/125Y, Thorlabs, Newton, NJ, USA) to which a halogen light source (HL-2000, Ocean Optics, Ostfildern, Germany) is coupled. The fiber tip is then imaged on the camera and APS (Figure 2e), and the co-alignment of the camera image and the APS can be verified. The APS consists of a 4 × 8 avalanche-diode array (pixel size: 1.6 mm) and therefore offers 32 pixels.

Of these 32 pixels, only the four central pixels are used for the tracking of a single speckle in order to maximize the acquisition rate.

**Figure 2.** (**a**) Experimental setup (transmission mode) for the remote photoacoustic measurements using the diode array sensing system. The imaging system can be moved mechanically in lateral directions, which is indicated by the red marks. (**b**) Picture of the sensing system. (**c**) The magnification of the sensing system is calculated at 5 using a USAF 1951 Test Target. (**d**,**e**) The camera position of the sensing system is calibrated using a multi-mode fiber. The diode signals and the camera image are shown when an illuminated fiber is placed in their center. Figures after [20].

The speckles are generated by cw illumination (532 nm, 80 mW, diameter: 0.75 mm) of the sample surface and imaged at *Z* = 20 cm. First, a convenient speckle is found by manually moving the imaging system using mechanical stages and visually tracking the camera image. A speckle is considered as convenient for the measurement if it is in the center of the camera and therefore in the center of the diode array (see Figure 3). Furthermore, the speckle size needs to be in the range of the pixel size of the diode array (1.6 mm). This is ensured by comparing the tracked speckle size to the inner circle (diameter 1.4 mm) of the illustrated target in Figure 3, which can be displayed by the camera software.

For the photoacoustic measurements, the samples are excited with a short laser pulse (Q-Smart 450, Quantel laser, Les Ulis, France). The laser parameters are as follows: *λ* = 1064 nm, pulse duration 5 ns, beam diameter 7 mm, pulse energy 90 mJ. These parameters result in an exposure of 230 mJ cm<sup>2</sup> , which is above the maximum permitted exposure (MPE) for single pulse excitation at 1064 nm (100 <sup>J</sup> cm<sup>2</sup> ). This, however, is desired to achieve a high signal amplitude for the demonstrated proof-of-concept experiments. The laser pulse triggers the acquisition start of the imaging system with a sampling rate of 8 MHz.

**Figure 3.** The camera image of an exemplary speckle used for photoacoustic measurement is shown. The speckle is placed in the center of the reference camera (blue target) and thus in the center of four measurements diodes P1/P2/P3/P4 (red) by manually moving the complete imaging system.

In order to prove the safe applicability of the sensing system, ex vivo skin measurements are performed with a total exposure below the MPE. This is achieved by replacing the described cw laser with a temporally pulsed laser diode (IBEAM-SMART-405-S-HP, Toptica Photonics, Gräfelfing, Germany) and by reducing the photoacoustic excitation energy. The laser diode pulse (*λ* = 405 nm) is temporally triggered by the short laser pulse for photoacoustic excitation and illuminates the tissue only for a duration of 30 μs after photoacoustic excitation at a peak power of 35 mW with an illumination radius of 300 μm at the tissue surface. This temporal speckle illumination together with a reduced excitation energy of 35 mW results in a total exposure that is below the MPE for soft tissue [21].

#### 2.2.2. Fiber-Based Single Speckle Sensing

Figure 4 shows the imaging unit and setup for the fiber based approach and camera images of a USAF 1951 Test Target and a selected speckle.

**Figure 4.** Experimental setup and imaging unit for the remote photoacoustic measurements using the fiber based approach (**a**). The proximal fiber bundle end can be moved in lateral directions, which is indicated by the red marks. The magnification for the camera arm is determined at 50 using a USAF 1951 Test Target (**b**). A convenient speckle that is centralized inside the photodiode measurement area is illustrated (**c**).

In contrast to the free-space approach, an imaging fiber bundle (30,000 fibers, imaging resolution 1 μm, working distance 30 μm, field of view diameter 240 μm) that can be used in endoscopy is used for speckle pattern imaging. At the proximal fiber bundle end, the speckles are imaged by the diode based imaging system mentioned in Section 2.2.1. In contrast to the free-space system previously described, a 20× magnification objective (WC95248318, Mitutoyo, Japan, 20×) is used. Furthermore, the imaging lens in the diode array arm is changed to *f* = 200 mm. Together with the magnification of the imaging fiber bundle (*M* = 2.5), these changes result in a magnification of *M* = 50 for the diode array arm and *M* = 25 for the camera arm. These higher optical magnifications allow speckle sensing at a near imaging distance of *Z* = 2 mm, which is desired for endoscopic usage.

The speckles are generated by focused cw illumination (532 nm, 100 mW, *Dill* = 50 μm) of the sample surface and imaged by the fiber bundle and APS system at *Z* = 2 mm. A convenient speckle is found and automatically centered inside the measurement area of the four photodiodes by analyzing the camera image and moving the proximal fiber bundle end in lateral directions. For the photoacoustic measurements, the phantoms are excited with a short laser pulse (Q-Smart 450, Quantel laser, Les Ulis, France), and the laser parameters are the following: *λ* = 1064 nm, pulse duration 5 ns, beam radius 7 mm, pulse energy 110 mJ. These parameters result in an exposure dose of 285 mJ cm2 , which is above the MPE for single pulse excitation at 1064 nm (100 <sup>J</sup> cm<sup>2</sup> ). This, however, is desired to achieve a high signal amplitude for the proof-of-concept experiments. The laser pulse triggers the acquisition start of the APS with a sampling rate of 8 MHz.

#### *2.3. Measurement Modes and Samples*

Using the described free-space and fiber based imaging systems, the samples are measured in transmission mode and reflection mode. For transmission mode, the photoacoustic excitation and speckle sensing take place at opposite sample sides, whereas they are on the same side for reflection mode.

The phantoms used in this work are made of the soft polymer PVCP (polyvinyl chloride plastisol; Standard Lure flex (medium), Lure Factors, Great Britain) and consist of two parts: absorber and scattering matrix. In order to adjust the optical properties for these parts, additives are used during the plastisol preparation process. A black plastic color changes the absorption coefficient *μa*, and TiO2-particles (titanium(IV)-oxide, Sigma Aldrich, Taufkirchen, Germany) adjust the reduced scattering coefficient *μ*- *<sup>s</sup>*. In this work, a color concentration of 7-vol-% and a TiO2-concentration of 4 mg mL(PVCP) is used for the absorbing and scattering phantom parts, respectively. The optical properties for these concentrations were determined at the excitation wavelength 1064 nm using spectrophotometric measurements and inverse adding doubling. The absorption coefficient for the absorbing phantom part is 106 cm−1, and the reduced scattering coefficient for the scattering part is 21 cm−1. The scattering coefficient for the absorbing part and the absorbing coefficient for the scattering part can be neglected.

For the measurements using the free-space setup, three different PVCP phantoms (PhAT1, PhAT2, PhAT3) with increasing distances between the absorber surface and detection surface (d) are measured in transmission mode. The "Ph" in the phantom name stands for phantom, "A" for the free-space setup, and "T" for transmission mode. Two PVCP phantoms are measured in reflection mode (PhAR4, PhAR5). The "R" in the sample name stands for reflection mode. Two skin tissue samples (skinAT1, skinAT2p) are measured in transmission mode. The "p" in the sample name stands for the pulsed speckle illumination, which ensures a total exposure below the MPE for the experiments with the skin tissue. The skin tissue was obtained from bisected pig heads, which were obtained from the local slaughterhouse (Unifleisch GmbH, Erlangen, Germany). Therefore, the approval of the Ethics Committee was not necessary. The tissue sample was prepared manually using a scalpel. For skinAT1, a PVCP-absorber was placed at the sample bottom by cutting out a hole in the tissue. For skinAT2, a PVCP-absorber was placed between a fat layer, which was obtained from a local supermarket, and a skin layer. In order to ensure good contact between the sample constituents, ultrasound gel was used, and for skinAT2p, a metallic sample holder gently pressed the tissues.

The speed of sound for the prepared skin sample was measured at 1300 <sup>m</sup> <sup>s</sup> and for the PVCP phantoms at 1330 <sup>m</sup> <sup>s</sup> with an ultrasound thickness measurement device (Mini Test 430, Elektro Physik, Germany). For each sample measured with the free-space setup, fifteen measurements were analyzed in order to ensure statistical relevance.

For the measurements using the fiber based imaging setup, two PVCP phantoms (PhBT1, PhBT2) were measured in transmission mode. The "B" in the sample name stands for the fiber based setup. The speed of sound for these samples were measured at 1349 <sup>m</sup> <sup>s</sup> with an ultrasound thickness measurement device (Mini Test 430, Elektro Physik, Germany). For these two phantoms, ten measurements were analyzed in order to ensure statistical relevance.

The density *ρ* of all PVCP phantoms was measured by the volume displacement of ethanol at 1200 kg <sup>m</sup>2s . The resulting acoustic impedance (*Zac* = *<sup>ρ</sup>c*) of the used phantoms in this work was therefore in the range of 1.60 <sup>×</sup> <sup>10</sup><sup>6</sup> kg m2s–1.62 <sup>×</sup> <sup>10</sup><sup>6</sup> kg m2s , which is in good agreement with the values of soft tissue: the impedance of fat tissue is 1.4 <sup>×</sup> <sup>10</sup><sup>6</sup> kg m2s and for muscle 1.62 <sup>×</sup> <sup>10</sup><sup>6</sup> kg m2s [22].

Figure 5 sketches the phantom position in regard to the excitation and illumination laser for the two measurement modes and shows the corresponding detection distances d of all samples. The absorber thickness is 3 mm for all samples, and d is varied by adjusting the thickness for the scattering sample part. For all samples, the mean detection times by automated single speckle analysis and their standard deviations are compared to the theoretical detection time.

**Figure 5.** The transmission setup (T) is sketched, and the phantom parts are marked (**a**). The reflection setup (R) is illustrated (**b**). Pictures of the two skin tissue samples are shown: skinAT1 with an absorber at the bottom (**c**) and skinAT2p with an absorber between skin and fat (**d**). The corresponding detection distances, theoretical detection times, and measurement modes are summarized in the table (**e**).

#### *2.4. Data Analysis for Single Speckle Analysis*

Figure 6 shows a speckle, which is initially in the center of the four measurement diodes P1, P2, P3, and P4 and moves to a different position due to the PA signal.

**Figure 6.** A perfectly round speckle, which is initially in the center of the measurement diodes is shown (**a**). The diode signals (P1, P2, P3, P4) can be used to compute its center of gravity *Csp*(*xsp*, *ysp*) and its total vector length *L* in order to detect speckle movements (**b**).

By using the temporal signals of the four diodes, it is possible to compute the temporal center of gravity of the speckle *Csp* (*xsp*, *ysp*) and its total vector length *L* by using Equations (2)–(4), similar to a four quadrant position sensitive diode [23]. Since the speckle moves if the surface tilts, this center of gravity is related to the photoacoustic signal, and it is possible to reconstruct the absorber depth *d*.

$$\propto\_{sp} = \frac{(P1 + P3) - (P2 + P4)}{(P1 + P2 + P3 + P4)} \tag{2}$$

$$y\_{sp} = \frac{(P1 + P2) - (P3 + P4)}{(P1 + P2 + P3 + P4)} \tag{3}$$

$$L = \sqrt{\mathbf{x}\_{sp}^2 + \mathbf{y}\_{sp}^2} \tag{4}$$

In order to prove the usability of the sensing system for remote photoacoustic detection, the detection times of *xsp*, *ysp*, and *L* are shown for the free-space approach. For *L*, the mean detection time *tmean* and its standard deviation *σ* are computed for all samples and verified to *ttheo* for the acoustic signal, which is calculated using d and c.

#### *2.5. Sensing Parameter Evaluation for Single Speckle Analysis*

#### 2.5.1. Sensitivity

The sensitivity in terms of minimal detectable tilt *Sd*,*<sup>α</sup>* is limited by the noise floor *σn f* of the PA measurements. The noise level *σn f* is computed by taking the standard deviation of a PA measurement data set before excitation. For this standard deviation computation, onehundred fifty data points before PA excitation are used. By determining the noise floor *σn f* and taking into account the relevant imaging parameters (*Z*, pixel size *dpx*, magnification *M*, illumination diameter *Dill*), *Sd*,*<sup>α</sup>* can be determined. The minimal detectable speckle shift *δs*,*min* is equal to the multiplication of *dpx* and *σn f* . By considering Equation (1), it is then possible to compute the minimal detectable tilt *Sd*,*α*: tan(*Sd*,*α*) = *<sup>δ</sup>s*,*min ZM* . Under the assumption that the investigated speckle pattern or single speckle consists of reflections from the complete illuminated surface area with the diameter *Dill*, the minimal detectable axial surface deformation is estimated by *Sd*,*nm* = tan(*Sd*,*α*) · *Dill*. The minimal detectable pressure *Sd* can be determined according to Equation (5) [24].

$$S\_d = \pi Z\_{ac} S\_{d, \text{sum}} f \tag{5}$$

The parameter *f* defines the frequency of the acoustic wave that should be detected, and *Zac* is the acoustic impedance. For the established system here, the maximum detectable frequency is half the frame rate. For the established free-space and fiber based sensing

system, the described sensitivities are computed and compared to the literature and to contact ultrasound transducers.

#### 2.5.2. Sensing Range

The sensing range, i.e., the tilt interval that is covered by the remote speckle analysis, is dependent on the sensitivity, speckle size, and sensor size. The sensitivity defines the minimal detectable tilt as treated in the previous section. The maximal detectable tilt *αmax* is defined by the sensor and speckle size. The sensing range for the single speckle analysis is discussed.

#### 2.5.3. Linearity

Simulations were carried out in order to evaluate the linearity and robustness against neighboring speckles, meaning speckles that are not situated inside the original image, but appear in the shifted speckle image. Four speckle patterns with a centralized speckle were analyzed. Figure 7 shows these speckle patterns. The images are moved in the horizontal direction with a shift amplitude of −0.26 to 0.26 (*xsp*,*real*) of the diode pixel size with a shift resolution of 0.025. For each shifted image, the horizontal center of gravity *xsp* for the fixed sensing diodes region, which is indicated in Figure 7, is computed according to Equation (2). The values of *xsp* are computed for all shifts and speckle images, compared to *xsp*,*real* and compared to linear behavior.

**Figure 7.** Speckle images (**A**–**D**) that are evaluated for their single speckle sensing capability. The diode sensing regions are indicated by the red rectangles.

#### **3. Results and Discussion**

#### *3.1. Photoacoustic Measurements*

Figure 8 shows measurement results in transmission mode of the three phantoms PhAT1, PhAT2, and PhAT3. They illustrate the speckle vector length *L*, which represents the temporal vibration profile of the surface under investigation. The detection time of the first peak in these profiles is marked with a black circle, because this time point corresponds to the photoacoustic signal. The surface expansion after the photoacoustic excitation results in a surface tilt change and thus in a speckle movement, which can be seen in *L*. For the phantoms PhAT1-PhAT3, the acquisition times increase as expected with increasing acoustic travel distance *d*. Considering *tsp*, the acquisition time increases as follows: 3.13 μs, 3.63 μs, and 4.38 μs. By using the speed of sound (1330 <sup>m</sup> <sup>s</sup> ), the following acoustic travel distances are calculated: 4.163 mm, 4.828 mm, and 5.825 mm. Thus, for each phantom, the acquisition time of the photoacoustic signal by speckle analysis corresponds to the geometrical phantom dimensions (d: 4 mm, 5 mm, 6 mm), taking into account the phantom production uncertainty.

**Figure 8.** The speckle vector length *tsp* is illustrated for measurements of phantoms PhAT1–PhAT3. For the three samples, the detection times of the initial generated photoacoustic signal are noted, and the corresponding signal peaks are marked.

Table 1 summarizes *tmean* for all photoacoustic acquisition times of all samples considering *L*, their *σ*, and the theoretical acoustic transit time *ttheo*. For illustration purpose, these results are also plotted in Figure 9.

**Table 1.** The mean value *tmean* for all photoacoustic acquisition times of all samples considering, its standard deviation *σ*, and the theoretical acoustic transit time *ttheo* are listed.


It is clearly visible that the speckle analysis mean detection times increase with bigger phantom dimensions and that they match the acoustic transit times. The small differences between the mean and theoretical detection time can be explained with inaccuracies for the phantom manufacturing and measuring process. The low standard deviations prove the repeatability and the fact that single speckle sensing allows precise photoacoustic sensing compared to the previous high-speed camera experiments. The standard deviation is in the range of approximately 0.1 μs, which results in a precision of 0.13 mm considering the speed of sound.

Based on the repeatability and successful verification with the theoretical transit time *ttheo* of the transmission mode and reflection mode measurements, it can be concluded that free-space and fiber based single speckle sensing is a reliable technique for the photoacoustic detection on phantoms and on skin tissue samples. Furthermore, the less expensive

low-resolution diode sensor, in contrast to the previously used high-speed camera, reaches a high sampling rate of 8 MHz, which allows precise photoacoustic sensing.

**Figure 9.** Mean detection times (*tmean*) and their standard deviation *σ* for the photoacoustic measurements. The sample names can be explained as follows: Ph stands for a phantom, A for the free-space setup, B for the fiber based setup, T for transmission mode, R for reflection mode, and p for pulsed speckle illumination below the MPE. The theoretical transit time of the photoacoustic signal is used for verification.

#### *3.2. Sensing Parameters*

#### 3.2.1. Sensitivity

Table 2 summarizes the relevant parameters for the determination of *Sd*,*nm* and *Sd* for the established sensing systems. For the purpose of comparison, *f* is selected at 1 MHz and 4 MHz, which represents the maximal detectable acoustic frequency for the established sensing system in this work.


**Table 2.** Relevant parameters for the determination of *Sd*,*nm* and *Sd* for the established sensing systems.

With the systems developed in this work, axial deformations of approximately 5 nm can be detected, which results in a pressure sensitivity of approximately 20 kPa for a 1 MHz acoustic wave. Horstmann et al. reached the following sensitivity parameters with a fullfield speckle interferometry approach: *Sd*,nm = 1 nm and *Sd*,1MHz = 1.5 kPa with a sensing bandwidth of 80 MHz [10]. These values, however, were achieved for measurements on silicone, which has a lower impedance (0.94 <sup>×</sup> <sup>10</sup><sup>6</sup> kg m2s ) than PVCP, which results in low *Sd*. Piezoelectric contact transducers that are especially designed and optimized for broadband PA detection achieve high sensitivities, which are dependent on the size and detection bandwidth. For a detection of acoustic frequencies in the range of 10 MHz to 50 MHz with an element size of 30 mm2, *Sd* can be estimated to lie between 1.5 Pa and 3.5 Pa [25]. Arrays have a smaller active area per detector element and thus lower sensitivity. An optimized ultrasonic line array can have a sensitivity of 110 Pa for a single element [26]. These sensitivities are better than the sensitivity for the established sensing system in this work. However, as explained in the Introduction, the interferometric setup is more complicated than the speckle analysis applied in the present investigation, and in comparison to the transducer, the single speckle analysis is contact-free. In addition, the speckle analysis sensitivity might even be improved by the usage of smaller photodiodes, tighter focusing of the cw illumination, or new data analysis techniques.

#### 3.2.2. Sensing Range

The single speckle analysis tracks the center of gravity of a single speckle that is positioned in the center of the four measurement diodes (Equations (2) and (3)). In general, the maximal detectable center of gravity coordinates (*xsp*,*max*, *ysp*,*max*) are defined by the single speckle diameter *ls* and the diode size (*dpx*) by Equation (6). For a larger speckle shift in regards to the sensor center, the single speckle would not be on the sensor completely, which leads to measurement errors. The maximum allowable speckle diameter is defined by twice the diode size. When assuming a speckle diameter of the pixel size, the maximum detectable center of gravity coordinates (*xsp*,*max*, *ysp*,*max*) are thus half the diode size.

$$x / y\_{sp,max} = (1 - \frac{l\_s}{2d\_{px}})d\_{px} \tag{6}$$

Table 3 gives an overview of the maximal detectable tilts in the horizontal and vertical direction (*αmax*,*x*, *αmax*,*y*). For the single speckle analysis, *x*/*ysp*,*max* can be determined according to Equation (6) with *ls* = 0.6*dpx*.

**Table 3.** Overview of the sensor size and surface tilt sensing ranges for the single speckle sensing approach.


The detectable tilt interval can be defined at [55 <sup>×</sup> <sup>10</sup>−5◦ ; 0.0642◦] for the diode system. These intervals can be converted according to the computations from Table 2 to pressure interval [31.67 kPa; 3690 kPa] for a PVCP surface and a 1 MHz acoustic wave. These values result in a dynamic range factor for the pressure detection of approximately 116, which is convenient for PA sensing.

#### 3.2.3. Linearity

Figure 10 illustrates the computed value for *xsp* for all shifts and speckle images and compares the results to linear behavior.

**Figure 10.** The computed values for *xsp* are shown over the *xsp*,*real* without (**a**) and with zero offset correction (**b**).

Though we tried to centralize the speckle inside the camera image, *xsp* is not zero for a zero shift (see Figure 10a). This can be explained by the speckle surrounding signal, which also falls into the diode sensing regions. Due to this effect, there is an offset for the computed *xsp* compared to linear behavior. This offset can be corrected, although there is still a clear difference between the computed *xsp* and linear behavior (see Figure 10b), which depends on the central speckle size and intensity distribution. Furthermore, the connection between *xsp*,*real* and *xsp* might be surjective. This means that there can be multiple values of *xsp* for one *xsp*,*real*. This effect occurs for Speckle Image B, as neighboring speckles that are shifted into the diode sensing region lower the signal strongly for higher shift magnitudes. The non-linearity problem can be corrected since each extracted shift can be mapped to a real shift when assuming a non-surjective behavior for the connection between *xsp* and *xsp*,*real*. This could be assured by an automated speckle finding software that analyzes the camera image and potential values for *xsp*,*real* and identifies a suitable speckle automatically.

#### **4. Conclusions and Outlook**

In previous studies, we already demonstrated the feasibility of remote photoacoustic sensing by multiple speckles analysis [16–18,20]. However, in these previous works, expensive detector systems were used and multiple speckles were analyzed, which limits the achievable sensing rate and thus also the resolution. Furthermore, the detection systems used were not fiber based and thus not suitable for endoscopic applications.

This study reports on a new, purely optical, non-interferometric modality for PA signal acquisition in the MHz range by analyzing a single speckle with four diodes and demonstrates its suitability for endoscopy. Based on the repeatability and successful verification of the transmission mode and reflection mode measurements, it can be concluded that a single speckle provides the information required for reliable PA detection on phantoms that mimic the optical and mechanical properties of tissue and skin samples. The successful fiber based measurements demonstrate the usability of the approach for endoscopic applications. These results are essential steps toward the future application of the technique in a potential imaging device or as a smart feedback system for laser procedures.

However, several challenge arise for the implementation of a future imaging system. Higher sensing rates need to be achieved in order to provide better sensing resolutions. For this work, the frame rate was limited to 8 MHz by the APS, which could be replaced by an economical four quadrant position sensitive diode [27]. With appropriate hardware, this PSD allows higher acquisition rates and thus better sensing resolutions. An array of

four quadrant diodes could be developed, and a measurement would be started if speckles were in a convenient position. Therefore, this sensor array would make a positioning unit obsolete and lower the costs of the sensing system significantly since a standard camera could be used for speckle position and shape tracking. Furthermore, the sensing sensitivity needs to be improved, which could be achieved by new hardware and a more precise speckle shift extraction algorithm [23]. This improved sensing sensitivity would reduce the required PA excitation exposure, which was above the MPE for soft tissue in most of the experiments for this investigation. A scan pattern of 10 × 10, which is considered to be sufficient for imaging, and a single photoacoustic measurement time of 10 μs result in a total measurement time of 1 ms per image, thus allowing imaging rates of 1 kHz. These mentioned steps together with appropriate reconstruction algorithms will allow PA imaging in the future [28].

**Author Contributions:** Conceptualization, B.L. and F.K.; formal analysis, B.L.; investigation, B.L. and D.S.; methodology, B.L., M.H., and Z.Z.; supervision, M.S. (Michael Schmidt), Z.Z., and F.K.; visualization, B.L.; writing—original draft, B.L.; writing—review and editing, M.H., M.S. (Moritz Späth), D.S., M.W., S.J.R., M.S. (Michael Schmidt), Z.Z., and F.K. All authors read and agreed to the published version of the manuscript.

**Funding:** This work was funded by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation)—Project Number 397972545. In addition, funding of the Erlangen Graduate School in Advanced Optical Technologies (SAOT) by the DFG in the framework of the German excellence initiative is gratefully acknowledged.

**Institutional Review Board Statement:** Not applicable.

**Informed Consent Statement:** Not applicable.

**Data Availability Statement:** The data that supports the findings of the study are provided within the article.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **Abbreviations**

The following abbreviations are used in this manuscript:


#### **References**


*Letter*
