This invention proposes a method,
system, device, and storage medium for
drug susceptibility detection based on single-bacterial segmentation and
deep learning. The method includes: acquiring bacterial microscopic images and fusing their bright-field and
fluorescence channels to obtain a fused
microscopic image; delineating and masking single
bacteria in the fused image to form a single-bacterial
mask dataset; using this dataset for
supervised training based on a single-
cell segmentation framework, and iteratively optimizing to obtain an optimized segmentation model; using the optimized segmentation model to automatically segment bacterial images and adaptively adjust thresholds, extracting single-bacterial regions and performing
quality control to obtain a qualified single-bacterial image set; performing hierarchical multi-fold cross-validation on this image set, training multiple
deep learning models, and selecting a core model based on the average performance of the validation set; predicting the image of the target
bacterial strain using the core model and outputting the
drug susceptibility detection result. This invention can rapidly identify the
drug susceptibility status of
bacteria, improving detection accuracy and stability.