Related Experiment Video
Updated: Jun 15, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
2.7K
Deep Anatomical Federated Network (Dafne): An Open Client-Server Framework for Continuous, Collaborative Improvement
Francesco Santini1,2, Jakob Wasserthal2, Abramo Agosti3
1Basel Muscle MRI, Department of Biomedical Engineering, University of Basel, Basel, Switzerland.
Radiology. Artificial Intelligence
|April 16, 2025
Summary
Dafne, a deep anatomical federated network, enhances radiologic image segmentation accuracy through collaborative learning. This open-source system improves model performance over time, demonstrating strong generalizability for diverse imaging data.
Area of Science:
- Medical Imaging Analysis
- Artificial Intelligence in Radiology
- Machine Learning for Healthcare
Background:
- Semantic segmentation of radiologic images is crucial for accurate diagnosis.
- Existing deep learning models often require large, centralized datasets.
- Decentralized and collaborative approaches are needed to leverage diverse data while preserving privacy.
Purpose of the Study:
- To introduce and evaluate Dafne (deep anatomical federated network), a novel decentralized, collaborative deep learning system.
- To enable semantic segmentation of radiologic images using federated incremental learning.
- To provide a freely available, open-source framework for collaborative AI in medical imaging.
Main Methods:
- Dafne utilizes a client-server architecture with an advanced user interface for prediction refinement.
- Federated incremental learning is performed on the client-side, with updates integrated into a central model.
- Evaluation involved local assessment on 38 lower leg MRI datasets and analysis of 639 real-world use cases.
Main Results:
- Dafne demonstrated statistically significant improvements in semantic segmentation accuracy over successive model generations (average Dice coefficient increase of 0.007 per generation, P < .001).
- Qualitative analysis showed enhanced model performance on various radiologic image types, including unseen data, indicating good generalizability.
- Real-world usage statistics confirmed the system's effectiveness and collaborative potential.
Conclusions:
- Dafne effectively improves radiologic image segmentation quality through a decentralized, collaborative federated learning approach.
- The system exhibits strong potential for continuous learning and generalization across diverse imaging datasets.
- Dafne offers a valuable, open-source tool for advancing AI-driven medical image analysis.
