Deep learning algorithms have been applied to digital image colorimetry (DIC) to improve color measurement and analysis, enabling the identification and quantification of analytes. In this work, we developed a smartphone-assisted platform for the detection of hypoxanthine (Hx) based on a ratiometric fluorescent probe (RF-probe) and deep learning algorithm for the first time with advantages of high sensitivity, in-field and simple operation, and accurate detection. Through an enzymatic cascade reaction, xanthine oxidase (XOD) degraded Hx to uric acid (UA) and H2O2, o-phenylenediamine (OPD) was converted to 2,3-diaminophenoxazine (DAP) by catalytic oxidation of H2O2 and OPD under horseradish peroxidase (HRP). The presence of DAP led to fluorescence quenching of BCNO quantum dots (BCNO QDs) at 440 nm and fluorescence enhancement of DAP at 570 nm due to the internal filtration effect (IFE). With the addition of Hx, the fluorescence color changed from blue to yellow. A good linear relationship was between I570/I440 and the Hx concentration in the range of 3-100 µM with a detection limit of 50 nM. Fluorescent color images of solution and test strips were acquired and extracted with a smartphone of our designed miniature device, which could be used to establish DIC for the detection of Hx with the linear relationship ranging from 0-100 μM and 0-40 μM, respectively. Then YOLOv5-assisted deep learning was applied to recognize and deal with the images, which offered improved accuracy, faster inference speed, flexibility, and efficient training. Therefore, our simple DIC platform was established for the in-field detection of Hx in real meat samples which was significant for the analysis of meat freshness and provided broad application potential for the analysis of other targets in food safety.
Read full abstract