AI-Enhanced Lateral Flow Assay Enables 3-Minute Quantitative Detection with Laboratory-Grade Accuracy

Abstract

Lateral flow immunoassay (LFA) remains one of the most widely used point-of-care testing (POCT) platforms for disease diagnosis, food safety assessment, and environmental monitoring. However, traditional LFAs typically require up to 30 min and offer only qualitative results, limiting their application where precise and rapid quantification is needed. In this study, we propose a Rapid and Accurate Deep Learning-Based Quantitative Lateral Flow Assay (RAD-LFA), specifically designed to overcome the inherent limitations of conventional LFA techniques. RAD-LFA integrates a Residual Network (ResNet) module for spatial feature extraction and a DyFormer module for dynamic temporal modeling, enabling precise quantification of target analytes within the first 3 min of the assay. This integrated framework significantly reduces detection time and enhances quantification accuracy, as demonstrated through extensive validation on Coronavirus disease 2019 (COVID-19) and hepatitis B virus (HBV) data sets. Experimental results show that RAD-LFA improves qualitative detection accuracy by 15% over expert visual interpretation and achieves robust quantitative performance with a coefficient of determination (<i>R</i>²) of 0.97. In clinical blind tests, RAD-LFA demonstrated excellent diagnostic performance, achieving 94% overall accuracy, 95% sensitivity, 92% specificity, and an <i>R</i>² of 0.9985 while markedly reducing the analysis time. Overall, RAD-LFA represents a promising, portable, and highly reliable POCT solution that effectively bridges the gap between decentralized testing and laboratory-grade quantification.

Ähnliche Arbeiten