Mammography Three clinics train a model without centralizing patient images.
Partner project by Innowise
Project profile
- CustomerMammography
- PartnerInnowise
- IndustryHealthcare
01
Challenge
Clinics wanted to learn from more varied imaging datasets while keeping raw patient images within each institution.
02
Solution
Collaborative AI training with locally retained images
Mask R-CNN models trained locally, with a central process aggregating updates. Preprocessing and evaluation followed a shared protocol.
Technologies
- Federated Learning
- Mask R-CNN
03
Results
Innowise reports segmentation AP gains of up to 68.6% against the weakest single-site baseline. This measures model performance, not an equivalent improvement in clinical diagnoses.
Project profile
- CustomerMammography
- PartnerInnowise
- IndustryHealthcare
Technologies
- Federated Learning
- Mask R-CNN
