Pathology foundation models for training-free case retrieval in salivary gland tumors
Here we explore whether content-based retrieval with current pathology foundation models facilitates rare cancer diagnostics without any task-specific training.
ProofPaper ↗
Key points
- In pathology, pattern recognition relies on a frame of reference built through extensive training and experience, which may still fall short when encountering rare cancer subtypes.
- Archives of previously diagnosed tumors often contain relevant precedents, yet identifying the most relevant cases remains challenging.
- Pathologists can query a digitized archive using a region of interest (ROI) from a whole slide image (WSI) and retrieve morphologically similar cases, along with their ICD-O codes and corresponding WSIs.
- We evaluate our approach by leave-one-case-out on an imbalanced dataset of 137 salivary gland tumor cases spanning 12 ICD-O morphology categories, comparing four pathology foundation models (Atlas2, Atlas2-B, Atlas2-S, and UNI2h).
Sources (1)
- [1]Pathology foundation models for training-free case retrieval in salivary gland tumorsarXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 15, 03:40 PM
Here we explore whether content-based retrieval with current pathology foundation models facilitates rare cancer diagnostics without any task-specific training.
In pathology, pattern recognition relies on a frame of reference built through extensive training and experience, which may still fall short when encountering rare cancer subtypes.
Extractive summary: sentences quoted from the sources.