A smartphone-integrated deep learning strategy-assisted rapid detection system for monitoring dual-modal...

This study focuses on the integration of a custom-built and optimally trained YOLO v5 model into a smartphone app developed with Java language. A dual-modal immunochromatographic rapid detection system based on a deep learning strategy for smartphones was developed for grade determination and predicting the concentration of aflatoxin B1 (AFB1). Innovative distance-type quantum dot microsphere fluorescent immunochromatographic chips enable semi-quantitative analysis by naked eye, and conventional colloidal gold nanoparticle colorimetric strips were also prepared. The compact and versatile hardware device making it easily integrable into smartphones of varying dimensions. Moreover, the wireless charging functionality of smartphones was to tackle power supply challenges. After optimized training, the accuracy, mAP@0.5, precision, and recall metrics of the YOLO v5 model all soared to 98 %. For the dual-modal immunochromatographic chips, the R² values for the standard curve fits were as high as 0.993, with a broad linear range of 0.05-40 ng/mL and a standard deviation lower than 0.03 at each concentration. Finally, this system determined the grade of the AFB1 concentration with an accuracy of up to 98 % and it exhibited an ultra-sensitive quantitative detection capability with a limit of detection as low as 2.2 pg/mL, showcasing the reliability of the deep learning strategy for practical applications in smartphones. This robust technological foundation paves the way for potentially community-based, family-oriented, and personalized applications.