Multimatrix Quant^AI-LFIA: An On-Device Deep-Learning Biosensing Platform with Charge-Transfer-Active Labels for...

Commercial gold nanoparticle-based lateral flow immunoassays (LFIAs) are widely used for point-of-care (POCT) diagnostics; however, inefficient signal transduction and the absence of intelligent interpretation result in low sensitivity and poor quantification. Here, we present an on-device platform (Multimatrix Quant^AI-LFIA) that integrates deep learning for the intelligent quantitative detection of staphylococcal enterotoxin B (SEB) across multiple biological matrices. By employing Fe-polydopamine (Fe-PDA) as a charge-transfer-active label for signal transduction, this system displayed superior colorimetric performance owing to efficient ligand-metal charge transfer (LMCT) transitions of Fe(III), achieving an initial visual limit of detection (LOD) of 5 ng/mL within 15 min. Further, the deep learning model enabled robust quantitative analysis under varying illumination conditions, extending the dynamic range to 0.1-800 ng/mL and reducing the LOD to 0.1 ng/mL, which was 50-fold lower than the visual readout. Through transfer learning, the model demonstrated reliable adaptation to serum and urine samples, yielding strong correlations between predicted and actual SEB concentrations (R² > 0.90). In blind tests, the Multimatrix Quant^AI-LFIA achieved an area under the curve (AUC) of 0.924 via a self-developed smartphone application, substantially outperforming naked-eye interpretation (0.700). This on-device platform provides rapid, accurate, and matrix-adaptive quantification, offering a practical pathway toward next-generation POCTs.