1. Introduction
Lateral flow assays (LFAs; aka. rapid tests) are inexpensive paper-based devices for rapid and specific detection of analyte of interest (e.g., COVID virus) in fluidic samples. LFA tests can be performed on-site by non-specialized users, and are significantly cheaper than laboratory-based alternatives [
1]. Because of these benefits, LFA tests are used in a wide range of applications, including agriculture [
2], water safety [
3], and point-of-care medical testing [
4], including most recently, COVID-19 testing [
5]. Consistent with widespread use of rapid tests, the global market of LFA industry is projected to reach a value of
$12.6 billion USD by the year 2026 [
6].
Despite obvious advantages of LFAs, these low-cost and rapid tests suffer from key limitations, such as having far inferior limit of detection and sensitivity compared to standard laboratory tests. For example, in roadside testing for levels of THC (the principal psychoactive compound found in cannabis) in oral fluids, the limit-of-detection of commercialized LFAs is between 25–40 ng/mL, while the recommended legal per se limit is between 0–2 ng/mL among different countries [
7]. A similar situation holds for workplace safety screening of THC where the desired limits of detection of 4–10 ng/mL cannot be accommodated by existing LFAs. As such, determination of THC intoxication at roadside or workplace currently requires use of more invasive, time-consuming, and costly tests such as a blood test administered by specialized professionals.
There have been many efforts recently to retain the universal use and inexpensive attributes of the LFAs but increase their sensitivity and limit-of-detection. One such method proposed using LFAs containing fluorescent polymetric beads [
8] as well as gold nanoparticles (GNPs), allowing for fluorescent interrogation of the LFA for enhancing the limit of detection. The drawback to this method is that standard LFAs are not currently manufactured using such fluorescent beads, and, thus, significant changes would have to be made in their manufacturing, which would likely increase their cost.
An alternative method, which requires no change in LFA manufacturing, is photothermal radiometry (PTR), wherein a focused laser light is used to excite the gold nanoparticles in a standard LFA, and the resulting thermal radiation is measured using an infrared camera [
9]. Our recent work show that this sensing method can increase the sensitivity and limit of detection of LFAs by over an order of magnitude [
10] while retaining the low-cost and rapid attributes of LFAs. A drawback with this method, however, is that it currently utilizes a thermal camera, which is prohibitively expensive for end-users, especially when LFAs are so highly valued for being inexpensive.
The purpose of this capstone project was to extend our recent innovations by integrating the sensing paradigm of photothermal radiometry in an Arduino-based system utilizing an inexpensive single-element far-infrared thermal sensor. Our results suggest that the significantly lower cost, end-user lab prototype significantly increases the sensitivity and limit of detection of LFA tests. As such, current work is expected to open the door for translation of our patented photothermal radiometry sensing platform to market as an end-user solution for significantly enhancing the detection performance of commercially available rapid tests.
2. Methods
2.1. LFA Principles
An LFA test is performed by depositing a controlled volume of fluidic specimen onto the sample pad. This fluid is then pulled by capillary action through the membrane to the conjugate release pad, which contains the antibodies specific to the type of analyte being tested for. In a competitive-style LFA test, such as the oral fluid THC test used in this study, the target analyte blocks the binding sites in the test line, allowing the antibodies to pass the test line and instead bind to the control line [
11]. In such an arrangement, spiking the LFA with specimens containing a concentration of analyte above the detection limit of the test (aka. a positive outcome), yields a test strip with non-existent test line and a visible control line. This can be seen in
Figure 1.
2.2. LFA Sensing via Photothermal Radiometry (PTR)
LFAs use GNPs as visible indicators in both the test and control lines. These GNPs selectively reflect ambient lighting, leading to colorimetric signals (normally dark purplish color) at test and control lines. The innovative PTR sensing method, however, takes advantage of selective absorption spectra of GNPs, meaning there are specific wavelengths at which light is absorbed very well by these GNPs but very weakly by the surrounding nitrocellulose paper [
12]. Accordingly, the PTR sensing method excites LFAs with a laser of specific wavelength to induce amplified thermal responses. Such responses can simultaneously be measured with infrared (IR) sensors and then quantified to yield signals that are proportional to the extent of accumulation of GNP-analyte pairs at LFA lines. To make PTR sensing method insensitive to varying ambient temperatures, or other sources of ambient thermal noise, the utilized laser excitation is intensity-modulated at a specific frequency (e.g., 1 Hz), creating thermal wave fields within the thickness of LFAs. Demodulation of acquired IR signals at the specific frequency enables removal of environmental thermal effects to enable very sensitive detection and quantification of light absorption signatures of the GNPs [
13].
2.3. Sensor Design Process and Final Design
The purpose of this capstone project was to demonstrate the feasibility of using inexpensive single-element IR sensors to produce results that were comparable to those produced by a thermal-camera-based system. To achieve this goal, an initial design was proposed that utilized an inexpensive single-element far-infrared sensor ($9; Wavgat Store, China) and 0.6 W laser diode ($20; 808-nm; Besram Technology Inc., Wuhan, China). Then, a parametric optimization design process was pursued. That is, one of the design variables (e.g., laser beam size) was changed, tests were performed on the control LFAs, the configuration performance was compared with previous configurations, and further configuration changes were then recommended until the design parameters were optimized.
The final optimized design composed of a hinged box with electronic components, consisting of Arduino-base controller, laser driver, servo driver, and power supply, housed below the testing deck. The device and system arrangement can be seen in
Figure 2. The testing deck holds a rack-and-pinion mechanism with a slot for insertion of an LFA cassette. This mechanism allowed for the LFA to be automatically moved so that the same laser-sensor pair could be used to interrogate both the test and control line. The laser was fixed at an optimal angle of incidence and distance to the LFA and controlled by the laser driver. The single-element far-infrared sensor mounted perpendicular to the line under test and was sampled by the Arduino-based controller.
2.4. Data Processing
During a test, datasets are collected separately for the test and control lines of an LFA. Both datasets contain a time series of radiometric measurements from the single-element sensor. Because of hardware limitations, the time series do not have perfectly spaced samples in time, so the series were then interpolated and resampled at regular time intervals to enable frequency domain analysis of the signals. A Fast Fourier Transform (FFT) was then performed on the resampled time series to transform the dataset to the frequency domain. The intensity of each response was then determined by comparing the response at the laser modulation frequency (1 Hz) to the respective noise floor. Analysis of the signals in the frequency domain via the FFT algorithm allows for effective rejection of signal noise, which is specifically essential when the signal-to-noise ratio is poor (i.e., test line color is faint). Finally, the response of the test line was normalized to the response of the control line. This normalized response (aka. amplitude metric) was found to be strongly correlated to the concentration of analyte in the fluid under test in the corresponding LFA.
Figure 3 shows representative interpolated and resamples time series and the corresponding frequency domain spectra for an LFA’s control and test lines. As expected, thermal-wave response is stronger for the control line in both time and frequency domains compared to those of the test line.
2.5. LFA Preparation/Samples
To test the performance of the developed system, ten LFAs were spiked with five known concentrations of THC (i.e., two spiked LFAs at each of the five concentrations). Spiking solutions were prepared by mixing a known volume of Delta-9 THC stock solution (Sigma Aldrich; Oakville, ON, Canada) with known volume of artificial saliva (Pickering Laboratories Inc.; Mountain View, CA, USA). Measured quantities of this solution were then serially diluted to obtain THC concentrations of 0 ng/mL, 2.5 ng/mL, 5 ng/mL, 10 ng/mL, 25 ng/mL. Ten LFAs (NarcoCheck; Montluçon, France), each with a cutoff concentration of 25 ng/mL, were then spiked using 200 uL of their respective solution and labelled accordingly.
3. Results and Discussion
Figure 4 depicts the mean ± STD amplitude metric data collected from the ten LFAs, each sampled ten times. Given the competitive-style design of LFAs, as THC concentration increases the test line fades out, leading to smaller amplitude metric signal. Results depicted in
Figure 4a show that for THC concentrations greater than 2.5 ng/mL, the amplitude metric of test line is significantly smaller than that of the LFA’s control line. As such, the limit of detection of developed end-user prototype was measured at 5 ng/mL.
Figure 4b shows the collected data from all THC concentrations tested. Qualitative assessment of the plot shows correlation of end-user output with THC concentration. To assess the data quantitatively, one-way ANOVA analysis of means was performed, which suggested presence of statistical significance (
p < 0.05). As such, the Tukey HSD post hoc pairwise comparison was conducted to identify which pairs of means were significantly different. As depicted in
Figure 4b, results suggest that end-user device can reliably resolve the THC concentration of the sample into one of four categories (<5 ng/mL, 5–10 ng/mL, 10–25 ng/mL, >25 ng/mL). Results of
Figure 4 show two major improvements: (1) the end-user device enhanced the detection limit from the nominal LFA threshold of 25 ng/mL to 5 ng/mL; (2) the end-user device enabled semi-quantitative detection of THC into four concentration categories (most of which are below the nominal detection threshold) as opposed to the conventional binary categorization of the LFA (i.e., <25 ng/mL OR >25 ng/mL). We would also like to note that more tests with larger sample sizes are required to further confirm the findings of this feasibility study.
Hardware costs for this prototype are found in
Table 1. This device is considerably less expensive than even the cheapest thermal camera-based LFA reader, which is about USD
$350 [
10]. With further development and optimization for mass manufacturing, the current USD
$51 cost using off-the-shelf components is expected to decrease significantly to further facilitate the translation of our innovation to the market as an end-user device.
This capstone project served as a feasibility study, thus, there is still much room for performance improvement. One possibility would be using a laser of a different wavelength at which GNPs are more absorptive. The current sliding mechanism is also prone to failure and a new method should be used to interrogate both lines simultaneously. More LFAs should also be prepared at a variety of different concentrations and at larger sample sizes for robust validation of end-user system performance.