Federated learning framework adapts AI models for fetal ultrasound analysis

Researchers have developed FedCC, a federated learning framework for localizing the corpus callosum in fetal ultrasound images. This approach is designed for low-resource clinical settings and avoids data sharing by adapting foundation models using lightweight modules. FedCC integrates a frozen DINOv2 backbone with a YOLO-based detection head and employs Low-Rank Adaptation (LoRA) to minimize computational and communication overhead. Evaluations on a multi-center dataset demonstrated FedCC's effectiveness, achieving a high mAP@50 of 0.857 and an F1-score of 0.803, while significantly reducing trainable parameters and communication costs compared to full fine-tuning. AI

IMPACT Enables privacy-preserving, low-resource AI deployment in medical imaging.

RANK_REASON Academic paper detailing a novel AI methodology. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• Alessandro Di Matteo
• DINOv2
• FedAvg
• FedCC
• Lora
• YOLO

• paper
• model release
• infra

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →