hatchmoment. scored by care · not by stars

clam-wsi-feature-extraction

B2-backed pipeline that tiles, segments, and embeds whole‑slide pathology images

notableTypeScript🧠 AI & ML

Whole‑slide images can be several gigabytes and explode into thousands of patches and large embedding tensors. This project wires OpenSlide, a truncated ResNet‑50 encoder, and a CLAM‑style workflow together with a Next.js dashboard and FastAPI backend, persisting raw slides, patches, and per‑slide embeddings directly to Backblaze B2 via the S3 API. It lets researchers and pathologists run the full extraction on a local CPU without needing massive local storage, and provides a UI to monitor cohorts and download artifacts. By handling storage, tiling, segmentation, and embedding in one cohesive, cloud‑native package, it simplifies a traditionally heavyweight pipeline.

b2-labsbackblaze-b2cloudstoragecomputational-pathologydigital-pathologyembeddingsfastapifeature-extractionhistopathologymedical-imagingmultiple-instance-learningnextjsobject-storageopenslidepythonpytorchs3sample-appwhole-slide-imaging
View on GitHub →

backblaze-b2-samples/clam-wsi-feature-extraction