Smartphone-Powered Automated Image Recognition Tool for Multianalyte Rapid Tests: Application to Infectious Diseases

Abstract

Point-of-Care Testing (POCT) is rapidly increasing, providing quick, user-friendly, and portable diagnostic tools. Lateral flow assays (LFAs) have been central to POCT, administering fast and cost-effective diagnosis. However, traditional LFAs are limited to qualitative or semiquantitative results. The integration of artificial intelligence (AI) and image analysis with LFAs has significantly improved diagnostic accuracy, result automation, and quantification where applicable. ΑΙ/image analysis algorithms are trained to automatically correlate the visual results with the presence of the analyte in the sample. Smartphone-based devices increase accessibility but also face challenges such as strip positioning and background lighting, which image analysis can potentially address. This study demonstrates a smartphone and machine vision-driven multicolor LFA, as well as an additional independent AI tool, for detecting pathogens like E. coli and SARS-CoV-2 in a single test. The developed system was successfully applied to real samples, providing accurate and multiplex results, advancing the field of infectious disease diagnostics. The results are presented as color, text, and audio messages, meeting all special needs of the users.

1

Principle of the multianalyte/multicolor rapid strip test. The targets are biotinylated and hybridized with a mixture of beads of different colors conjugated to the specific detection probes. SARS-CoV-2 detection probe is coupled to the beads through dA/dT hybridization using polydT-conjugated beads. The hybrids are captured by immobilized streptavidin at the strip’s test zone, forming a colored line depending on the target(s) present in the sample.

2

Optimization studies. Conjugation reaction. (a) Volume of EDC. (b) Amount of probe. (c) Composition of the running buffers. Different running buffers were tested in order to get a good flow of the beads and good clearness of the membrane of the strip with the most intense color (signal) at the test zone. 1: 2% Tween-20, 1× PBS pH 7.4; 2: 1× TE, 1% glycerol, 1% Tween-20; 3: 1% glycerol, 1% Triton X-100, 1× PBS pH 7.4; 4: 1% Tween-20, 1× PBS pH 7.4; 5: 2.5% Tween-20, 1× PBS pH 7.4; 6: 3% Tween-20, 1× PBS pH 7.4; 7: 0.1% Triton X-100, 0.05% SDS, 1× PBS pH 7.4; 8: 0.1% Tween-20, 0.05% SDS, 1× PBS pH 7.4; 9: 1× PBS pH 7.4; 10: 1% sucrose, 1× PBS pH 7.4; 11: 0.5% Triton X-100, 1× PBS pH 7.4; 12: 0.5% Tween-20, 0.5% sucrose, 1× PBS pH 7.4. (d) Volume of the stock of the beads. Deposition and immobilization of reagents on the membrane of the strip. (e) Time of UV irradiation. (f) Deposition speed for streptavidin. (g) Amount of immobilized streptavidin. (h) Amount of immobilized polydA probe. SA, streptavidin; N, negative; CZ, control zone; TZ, test zone.

3

(A) Detectability of the rapid test for all four targets. Different amounts (0–100 or 500 fmol) were analyzed by strip test for each target and the corresponding colored beads. Detectability study for with red beads, for with blue beads, for with green beads, and for SARS-CoV-2 with orange beads. CZ, control zone; TZ, test zone. (B). Detectability of the rapid test in the presence of all beads. Detectability study for E. coli in a mixture of beads, for a mixture of and in the presence of all beads and finally, a mixture of and in the presence of all beads.

4

Detection of the targets in a mixture of the four beads and specificity of the multicolor test. (a) Each target was hybridized separately with the mixture of the beads. Only the color corresponding to the target present in the sample was formed at the strip’s test zone, proving the multicolor test’s excellent specificity. (b) Simultaneous detection of two targets with the beads’ mixture. N, negative; SP, ; EC, ; HI, ; CZ, control zone; TZ, test zone.

5

Workflow of the mobile application. (a) Login screen. (b) Registration screen. (c) Initial test result history screen for nonadmin user. The user must first log in or sign up with their credentials. Then, the test result page appeared on the screen. In this screen, the user can make one of the three following decisions: (i) Logout: it disconnects the user from the application and returns them to the login screen; (ii) Navigate to the disease screen that displays defined by its name and the color of the test line of the strip, which categorizes the strip as positive. (iii) Navigate to the camera screen: (i) Green, which means positive to the disease; (ii) Red, which means negative to the disease; and (iii) Yellow, which indicates the infectious diseases identified by the system. An infectious disease is characterized by its name and the color of the test line of the strip that categorizes the strip as positive and (iii) Navigate to the camera screen: navigate to the camera screen to take a picture of the test strips. (d) Image capture of the test strips. (e) Successful photo upload. (f) Result screen with audio cue and image test results. (g) Updated test results screen. After taking a picture, the user will receive a confirmation message on the screen indicating that the image was sent successfully. If the test results are positive for an infectious disease, an audio cue will be played, and a screen will display a vertical list of the diseases for which the user has tested positive for. Finally, the user is directed to the test result history screen, which displays the updated information.

6

Confusion matrix. The actual classification of the samples was compared to that of the interpretation of the visual outcomes of the strip tests of all analyzed samples by the developed systems. (a) Image analysis-based system and (b) AI-based system.