Machine learning-assisted smartphone fluorescence sensing for ultrasensitive and point-of-care testing of acid...

Fluorescence-based methods are crucial in clinical diagnosis due to their high sensitivity and visualization capabilities. Traditional single-emission sensors often have the problems of background interference and poor visual recognition in complex biological environments. Although multicolor fluorescence visualization sensors can broaden the color gamut to improve resolution, they are still limited by subjective human color judgment and individual differences.

We developed a three-color fluorescence sensing platform that integrates smartphone machine learning (ML) to enable portable and ultra-sensitive acid phosphatase (ACP) detection. The sensing system consists of manganese dioxide nanosheets (MnO₂ NS), o-phenylenediamine (OPD), and red carbon dots (R-CDs). ACP triggers a cascade that generates ascorbic acid (AA) to reduce MnO₂ NS, resulting in a dual signal response, while R-CDs reverts to red fluorescence, achieving a pronounced colorimetric change in the solution from yellow to orange to purple. By capturing images of the solution using a smartphone and interpreting complex color features using ML models, we achieved accurate quantification of ACP in the range of 0.5-7.5 mU/mL. This platform features excellent selectivity and stability, demonstrating outstanding anti-matrix interference capability in complex human serum samples.

This study not only provides a novel and reliable method for the analysis of ACP activity, but also establishes a universal biomarker sensor. The platform expands intelligent point-of-care testing (POCT) systems for the development of other clinically relevant biomarkers, facilitating practical applications for POCT diagnosis and real-time biochemical monitoring.