Deep learning-enabled microfluidic digital PCR platform for efficient seven-color quantification

Digital PCR (dPCR), as a high-sensitivity technology for absolute nucleic acid quantification, holds significant value in biomedical research and environmental monitoring. However, current platforms still face challenges in multiplex fluorescence detection and rapid, high-precision droplet imaging. Moreover, the detection process is time-consuming (2-3 hours) and involves high costs. In this study, an integrated micro-droplet digital PCR (ddPCR) detection and analysis system was developed, featuring a droplet-based microfluidic chip, a high-precision thermal cycling module, and a seven-color filter-wheel-based imaging system (ATTO425 to CY7) to facilitate a seamless workflow from droplet generation to multiplex imaging. To address the challenges of identifying and segmenting massive droplets in complex fluorescence backgrounds, this paper proposes a detection method based on the You Only Look Once version 5 (YOLOv5) deep learning architecture. By integrating global coordinate remapping and sliding-window detection, the system enables rapid processing of ultra-high-resolution images (2448 × 10 000 pixels). The end-to-end analysis pipeline achieved 99.8% overall accuracy in under 800 ms. Consequently, the total detection cycle for the full digital PCR process has been successfully reduced to under one hour. Furthermore, full-process validation experiments demonstrated excellent linearity across all fluorescence channels, with R² values exceeding 0.999, and a coefficient of variation (CV) for quantitative repeatability of less than 2% across various concentrations. These results verify the system's precision, stability, and reproducibility. The developed system significantly enhances the throughput and accuracy of ddPCR detection, and the proposed algorithm further advances the practical application of deep learning in digital PCR image analysis.