*3.2. Experiments*

The schematic of our experimental setup is shown in Figure 4a. A tunable optical parametric oscillator (OPO) system (Phocus Mobile, Opotek, CA, USA) is pumped by a Nd; YAG laser (Brilliant B, Bigsky, MT, USA) is employed as the laser source. The laser works at the wavelength of 720 nm with pulse repetition rate of 10 Hz and pulse duration of 5.5 ns. A Verasonics system (Verasonics Inc., Kirkland, WA, USA) is controlled by a computer to receive raw PA data. An ultrasound linear array (L7-4, Philips, Amsterdam, The Netherlands) with 128 elements, 5.208 MHz center frequency and 0.298 mm element pitch was used. Ultrasonic coupling agen<sup>t</sup> is used as the acoustic transmission medium between phantom and ultrasound probe.

**Figure 4.** (**a**) Schematic of the experimental setup; (**b**) the phantom used as imaging target (two rubber threads); (**c**) the phantom to acquire the deconvolution kernel (one tungsten wire); (**d**) a photo of the experiment setup.

The phantoms used in this experiment were made of porcine gelatin with 8% concentration as shown in Figure 4b,c. Figure 4b is the phantom used as the imaging target in our experiment which embedded with two rubber threads of diameter 1 mm and Figure 4d shows the placement of phantom, probe and laser in this experiment. The arrows in Figure 4b,c both point to the position where a probe acquired a signal of each phantom while carrying out the experiment. During the process of signal acquisition, each set of signals contained approximately 200 frames for a continuous period to average.

The deconvolution kernel is acquired from the middle transducer element of the probe, which detected a single tungsten wire of diameter 0.3 mm (as Figure 4c); approximately the smallest size L7-4 probe can measure. The normalized deconvolution kernel (as shown in Figure 5), averaged from 200 frame signals, was used as the deconvolution kernel in this experiment.

**Figure 5.** Diagram of the normalized deconvolution kernel in the experiment.

The comparison of the reconstructed result is shown in Figure 6. We can observe that the two oblique rubber wires are too close to look separated in the image reconstructed with original PA signal, as shown in Figure 6a. Without knowing the imaging targets in advance, it is difficult to distinguish the structure visually. After applying our proposed method, as shown in Figure 6b, the two lines are completely separated, and the realistic inner structures of this phantom can be more clearly identified. Meanwhile, image resolution is enhanced, and the artifact is slightly reduced.

To quantify the width of gaps in Figure 6 and compared to the realistic width of gap in Figure 4b, we measured the width of gaps in Figure 6. In Figure 4b, the diameter of the two rubber wires in the phantom are both 1 mm, we can observe that the realistic gap, where the arrow mark is pointing to, is approximately 0.4 mm. In respect of the reconstructed image of the original signal as shown in Figure 6a, the width of the gap is about 0.08 mm. For the reconstructed image of the processed signal as shown in Figure 6b, the width of the gap is about 0.3 mm. From the quantification, we can find that the image of processed signal more aptly indicates the accurate width and position of realistic targets.

**Figure 6.** Reconstruction results of the experiment. (**a**) the reconstructed image of original signal; (**b**) the reconstructed image of processed signal.

#### **4. Discussion and Conclusions**

From the contrast of signals in the simulation before and after processing, the bipolar signals are converted into non-negative signals, whereas the processed signals in the experiment are relatively monopolar, but not strictly monopolar. The deconvolution result is sensitive to noise, which is inevitable in the signal acquisition procedure. Meanwhile, the relatively monopolar signals have some benefit in reducing artifacts in the final reconstructed image. Widespread artifacts as shown in Figure 2d appear around imaging targets when using the strictly monopolar signal, whereas artifacts can be slightly alleviated if the relatively monopolar signal are used.

In this work, we demonstrate an image optimization method by processing the raw PA signals using signal deconvolution and EMD. Using this method, raw bipolar signals are converted into quasi-monopolar signals and more detailed information can be obtained to alleviate the signal aliasing phenomenon. With the processed signals, PA image can be optimized, especially in resolution and edge sharpness. Moreover, the micro-structures, which were previously hidden and buried by large imaging targets, can now be revealed and separated. Both numeric simulations and phantom study illustrate the effectiveness of this method. As this method can help to distinguish between close objects and reveal detail information, it can potentially help with diagnoses, especially on tissues with micro-structures.

**Author Contributions:** Conceptualization, C.G. and J.Y.; data curation, C.G.; formal analysis, C.G.; funding acquisition, J.Y.; investigation, C.G.; methodology, C.G., J.Y. and Y.Z.; project administration, J.Y.; resources, J.Y., Q.C. and X.W.; software, C.G.; supervision, J.Y.; validation, C.G., Y.C. and Q.C.; visualization, C.G.; writing–original draft, C.G.; writing–review and editing, J.Y. and Y.Z.

**Funding:** This research was funded by the National Key Research and Development Program of China (No. 2017YFC0111402) and the Natural Science Funds of Jiangsu Province of China (No. BK20181256).

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