Rapid and ultra-sensitive detection of foodborne pathogens by deep learning-enhanced microfluidic biosensing

Rapid and sensitive detection of foodborne pathogens is essential for ensuring food safety and protecting public health. In this study, we developed an innovative microfluidic fluorescence digital analysis platform enhanced by deep learning to detect pathogens at ultra-low concentrations. The biosensor features a staggered herringbone double-spiral (SHDS) microfluidic design, seamlessly integrating bacteria capture, detection, and release processes using Quantum dot (QD)-Aptamer conjugates for precise identification. Fluorescence image analysis, powered by a Resnet-18-based convolutional neural networks (CNN), directly quantifies Escherichia coli (E. coli) concentrations from fluorescence images, streamlining data processing and increasing sensitivity. The platform offers a linear detection range from 10 to 3 × 10⁶ CFU/mL (R² = 0.990), achieves capture efficiencies of up to 100 % at low bacterial concentrations (4 × 10² CFU/mL), and offers an ultra-low detection limit of 2 CFU/mL within just 1.5 hours. The CNN model effectively filters out background noise and interferences, achieving over 99 % predictive accuracy. Validation using milk and chicken samples resulted in high recovery rates (96.7 % to 104.0 %). This biosensor presents a rapid, reliable, and practical solution for pathogen detection in complex food matrices, significantly improving food safety and security. • Developed a novel microfluidic biosensor integrating E. coli enrichment, detection, and release. • Achieved up to 100 % E. coli capture efficiency at 4 × 10² CFU/mL using asymmetric herringbone structures and Con A-modified channels. • Enhanced fluorescence analysis with deep learning , achieving a detection limit of 2 CFU/mL. • Demonstrated high predictive accuracy (>99 %) and recovery rates (96.7 % to 104.0 %) in food samples. • Applied a CNN model to eliminate background noise, enabling precise detection in complex food matrices.