Latest research publication

Excited to share our latest research publication! Our new paper, “Performance of federated versus centralized learning for mammography classification across film–digital domain shift”, is now published in Frontiers in Digital Health.

What’s it about? Deep learning in mammography relies on large, diverse datasets — but clinical data often remain siloed. Federated Learning (FL) offers a privacy-preserving alternative by enabling collaborative model training without sharing raw data. But how well does FL perform when imaging data come from very different domains, such as scanned film vs. digital mammography?

Key insights from our study:

• FL performs on par with centralized learning when all data come from similar domains.
• Under strong film–digital domain shift, FL maintains high performance on digital images but struggles on film-based data, showing reduced precision.
• Popular FL variants (FedAvg, FedProx, SCAFFOLD, FedBN) do not fully overcome this domain mismatch.
• Increasing image resolution helps but cannot close the performance gap.
• The findings highlight the need for domain-aware and personalized FL approaches to ensure safe, reliable deployment in breast imaging.

Why this matters: As healthcare moves toward privacy-preserving AI, understanding the limits of federated learning is crucial — especially in high-stakes applications like breast cancer detection.

Read the full article HERE