Swarm learning for decentralized artificial intelligence in cancer histopathology
Problem
Training robust AI on medical images needs large multicentric datasets that face practical, ethical and legal obstacles to sharing.
Why it matters
Predicting molecular alterations from routine slides could improve colorectal cancer diagnosis without extra testing.
What the AI did
Swarm learning let partners jointly train AI models on over 5000 patients' gigapixel histopathology images without data transfer; models predicted BRAF mutational status and MSI from H&E slides.
Results
AI models trained via swarm learning predicted BRAF mutational status and MSI from colorectal cancer H&E slides across three cohorts (>5000 patients); accuracy metrics not stated in source.
Impact
First demonstration (per abstract) of swarm learning enabling decentralized multicentric AI training in histopathology.
Record
| Venue | Nature Medicine |
|---|---|
| Published | 2022 |
| Work type | article |
| Authors | Oliver Lester Saldanha, Philip Quirke, Nicholas Paul West, Jacqueline A. James, Maurice Bernard Loughrey, Heike Irmgard Grabsch, Manuel Salto‐Tellez, Elizabeth Alwers et al. |
| Citations | 224 (OpenAlex, at import time) |
| References | 59 |
| Model or method | not stated in source |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1038/s41591-022-01768-5 |
| Open access | hybrid |
| Keywords | swarm learning, histopathology, colorectal cancer, MSI, BRAF, decentralized AI |
Importance rating
significant — thousands to millions affected
Addresses multicentric medical AI training obstacles via swarm learning on >5000 patients' histopathology images, predicting BRAF status and MSI in colorectal cancer without data transfer.