Fast and scalable search of whole-slide images via self-supervised deep learning
Problem
Searching large repositories of gigapixel whole-slide images for morphologically similar slides was too slow because retrieval speed scaled with repository size.
Why it matters
Fast similar-case retrieval in digital pathology supports clinical diagnosis and large-scale research.
What the AI did
SISH uses self-supervised deep learning to encode slides into discrete latent representations with a tree data structure for fast search and uncertainty-based ranking, needing only slide-level labels.
Results
WSI retrieval speeds independent of repository size; evaluated on multiple tasks (specifics truncated)
Impact
Open-source, peer-reviewed algorithm enabling scalable similar-slide search in pathology archives.
Record
| Venue | Nature Biomedical Engineering |
|---|---|
| Published | 2022 |
| Work type | article |
| Authors | Chengkuan Chen, Ming Y. Lu, Drew F. K. Williamson, Tiffany Chen, Andrew J. Schaumberg, Faisal Mahmood |
| Citations | 126 (OpenAlex, at import time) |
| References | 60 |
| Model or method | SISH (self-supervised deep learning) |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1038/s41551-022-00929-8 |
| Open access | hybrid |
| Keywords | whole-slide images, self-supervised learning, digital pathology, image retrieval, histology |
Importance rating
significant — thousands to millions affected
Enables fast similar-slide retrieval in large pathology archives, aiding diagnosis including rare cancers; evaluated on over 22,000 patient cases and 56 disease subtypes—clear clinical and research benefit.