AtlasPatch: Efficient Tissue Detection and High-throughput Patch Extraction for Computational Pathology at Scale
New open-source framework uses a Segment-Anything (SAM) model to process 30,000 whole-slide images with 98.6% precision.
Deep Dive
A research team led by Mahdi S. Hosseini built AtlasPatch, an open-source framework for computational pathology. It couples a foundation-model tissue detector (trained on ~30,000 WSI thumbnails) with high-throughput patch extraction. The tool achieves 0.986 precision and reduces end-to-end WSI preprocessing time by up to 16x versus standard pipelines. Pathologists can use it for quality control, while AI researchers can stream patches directly into feature-extraction models for faster training.
Why It Matters
Removes a major bottleneck for scaling AI in medical diagnostics, enabling faster analysis of large patient cohorts.