Related Experiment Video
Updated: Jul 2, 2025
04:17
DNA Virus Detection System Based on RPA-CRISPR/Cas12a-SPM and Deep Learning
Published on: May 10, 2024
753
A comparative study of federated learning methods for COVID-19 detection
Erfan Darzi1, Nanna M Sijtsema2,3, P M A van Ooijen2,3
1Harvard Medical school, Harvard University, 300 Longwood avenue, Boston, United States. e.darzidehkalani@rug.nl.
Scientific Reports
|February 16, 2024
Summary
Federated learning (FL) enables COVID-19 detection model training across hospitals without sharing private data. The Cyclic Weight Transfer algorithm shows superior performance and resource efficiency, especially with fewer participating hospitals.
Area of Science:
- Artificial Intelligence
- Medical Imaging
- Computer Science
Background:
- Deep learning models for COVID-19 diagnosis require large datasets, but data sharing is limited by privacy concerns.
- Federated learning (FL) offers a privacy-preserving approach for training AI models across multiple institutions.
Purpose of the Study:
- To evaluate the performance and resource efficiency of five FL algorithms for COVID-19 detection using CNNs.
- To identify optimal FL strategies for decentralized medical image analysis.
Main Methods:
- Assessed five FL algorithms for COVID-19 detection with Convolutional Neural Networks (CNNs).
- Varied parameters including the number of participating entities, federated rounds, and selection algorithms.
- Analyzed performance and resource utilization in a decentralized setting.
Main Results:
- The Cyclic Weight Transfer algorithm demonstrated superior performance in COVID-19 detection.
- This algorithm was particularly effective when the number of participating hospitals was limited.
- Resource efficiency was also a key consideration in the evaluation.
Conclusions:
- Federated learning is a viable solution for privacy-preserving AI in medical diagnostics.
- The Cyclic Weight Transfer algorithm presents a promising approach for efficient FL in COVID-19 detection.
- Findings have implications for deploying FL in medical imaging and other decentralized AI applications.
