Deep learning-driven multiplexed mycotoxin detection via color-size encoded microbead imaging

Mycotoxins pose severe threats to food safety and public health, especially with synergistic toxicity from co-contamination. Herein, a deep learning-driven two-dimensional (2D)-encoded single microbead imaging decoding platform is developed for homogeneous analysis of four mycotoxins. The aptamer-recognition-triggered DNAzyme walker-hybridization chain reaction (Dz-HCR) cascade design enables direct fluorescent signal lighting and amplification on microbeads, avoiding complex separation and signal probe preparation. By leveraging two different-sized microbeads and two fluorophores modified at HCR hairpins, an ingenious color-size 2D-encoding strategy is established for high-throughput analysis. The YOLOv11 deep learning model enables rapid, accurate decoding and analysis of fluorescence images with multi-dimensional information, improving data processing efficiency. This platform exhibits high sensitivity (<1 pg/mL), wide linear ranges, and excellent specificity. The desirable spiked recovery (85.6%-114.0%) in maize and plant-based meat analogs and the consistency with HPLC-MS/MS confirm practical applicability. This method provides a promising strategy for high-throughput mycotoxin detection in food safety monitoring.