Related Experiment Videos
FSCL-BC: Federated supervised contrastive learning for breast cancer diagnosis with high sensitivity
Faisal Ahmed1, David Sánchez2, Zouhair Haddi3
1Universitat Rovira i Virgili, Department of Computer Engineering and Mathematics, CYBERCAT-Center for Cybersecurity Research of Catalonia, ComSCIAM-Center for Computational Science and Applied Mathematics, Tarragona, Catalonia, Spain; NVISION Systems and Technologies SL, Barcelona, Spain.
Computer Methods and Programs in Biomedicine
|June 12, 2026
Summary
A new AI model, FSCL-BC, improves breast cancer diagnosis from ultrasound images by enhancing sensitivity and maintaining patient privacy. This federated learning approach allows collaborative training without data sharing, overcoming common challenges in AI development.
Area of Science:
- Artificial Intelligence in Medical Imaging
- Machine Learning for Healthcare
- Radiology and Diagnostic Imaging
Background:
- Accurate breast cancer diagnosis from ultrasound images is challenging due to data limitations and class imbalance.
- Developing robust AI models requires diverse data, which is difficult to obtain due to privacy and regulatory concerns.
- Current AI models struggle with generalization and high sensitivity, increasing the risk of missed cancer cases.
Purpose of the Study:
- To develop an accurate AI model for breast cancer prediction from ultrasound images.
- To enable collaborative AI model training across multiple hospitals without direct data sharing.
- To address privacy concerns, data ownership issues, and regulatory constraints in AI development for medical imaging.
Main Methods:
- Introduction of FSCL-BC, a privacy-preserving method integrating supervised contrastive learning within federated learning.
- Hospitals collaboratively train the AI model while keeping their data on-premises.
- Only model parameters trained on private data are exchanged, ensuring data privacy.
Main Results:
- FSCL-BC achieved substantially higher diagnostic performance, especially sensitivity, compared to centralized and vanilla federated training.
- FSCL-BC improved sensitivity by 11.5%, Youden's J index by 7.2%, and F1 score by 3.7% on average.
- The method provides intrinsic privacy protection in a realistic federated setting.
Conclusions:
- FSCL-BC offers improved diagnostic accuracy and enhanced patient privacy for AI models in breast cancer diagnosis.
- It is a promising and practical solution for real-world, resource-constrained settings.
- The federated learning approach facilitates the development of generalizable AI models without compromising patient data.
