Intelligent Handling of Noise in Federated Learning with Co-Training for Enhanced Diagnostic Precision

EasyChair Preprint 14772

Abstract

Federated learning (FL) is a decentralized machine learning approach that enables multiple distributed clients to train a model jointly while keeping their data private. However, in medical data, particularly images like brain MRIs, often suffer from noise like acquisition noise and scanner artifacts, which can lead to misdiagnoses. Existing methods for handling noise often rely on data transmission, increasing communication burden and privacy risks. We propose a novel Federated Learning (FL) approach with co-training framework. Our method leverages the inherent discrepancy in the learning ability of the local and global models in FL, which can be regarded as two complementary views. By iteratively exchanging samples with their high confident predictions, the two models “teach each other” to suppress the influence of noisy labels. Our method, tested with several aggregation strategies on a benchmark dataset of 1300 Brain MRIs, and our own collected data from Biobank UK and increased accuracy from 83.05% to 85.20%. This demonstrates the effectiveness of our proposed model for accurate and privacy-preserving medical image analysis. While the proposed model offers high accuracy, it demands more computational resources, making it better suited for powerful servers than personal devices.

Keyphrases: Diagnostic Precision, Federated Learning, Machine learning approach, Noise Handling, co-training

@booklet{EasyChair:14772,
  author    = {Farah Farid Babar and Faisal Jamil and Faiza Fareed Babar},
  title     = {Intelligent Handling of Noise in Federated Learning with Co-Training for Enhanced Diagnostic Precision},
  howpublished = {EasyChair Preprint 14772},
  year      = {EasyChair, 2024}}