DiffuPath: Diffusion-Guided Waterborne Pathogen Detection in Microscopy Images

Waterborne pathogens like Giardia lamblia, Cryptosporidium parvum and Vibrio cholerae have a severe public health hazard and require rapid and accurate microscopy-based detection. However, manual screening is time-consuming and operator-dependent, and automatic methods are currently suffering from high false positive rates and a lack of generalization to different conditions for imaging. To overcome these shortcomings, DiffuPath is introduced, which is a diffusion-guided framework with denoising prior combined with pathogen detection head and uncertainty estimation module for robust waterborne pathogen screening. The diffusion prior learns morphology-aware, noise-robust representations to suppress debris-induced false activations. Experiments on the CDC DPDx Parasite Image Library, which includes more than 2000 images with 18 species of pathogens, show that DiffuPath has achieved a Precision of 93.4%, Recall of 91.1%, F1-score of 92.2% and mAP@0.5 of 95.0% against YOLOv8, Faster R-CNN and RetinaNet as baselines with an inference latency of 18 ms/image. These results validate DiffuPath as an efficient solution for real-world water quality monitoring.