Non-parametric regularization for class imbalance federated medical image classification

TL;DR

This paper introduces a non-parametric regularization method for federated learning to improve medical image classification under severe class imbalance, demonstrating superior performance over existing approaches.

Contribution

The paper proposes FedNPR and FedNPR-Per, novel non-parametric regularization techniques that enhance feature discriminability in federated learning for medical imaging.

Findings

FedNPR outperforms state-of-the-art FL methods in skin lesion classification.

FedNPR improves intracranial hemorrhage identification accuracy.

Regularization enhances existing FL approaches' performance.

Abstract

Limited training data and severe class imbalance pose significant challenges to developing clinically robust deep learning models. Federated learning (FL) addresses the former by enabling different medical clients to collaboratively train a deep model without sharing privacy-sensitive data. However, class imbalance worsens due to variation in inter-client class distribution. We propose federated learning with non-parametric regularization (FedNPR and FedNPR-Per, a personalized version of FedNPR) to regularize the feature extractor and enhance useful and discriminative signal in the feature space. Our extensive experiments show that FedNPR outperform the existing state-of-the art FL approaches in class imbalance skin lesion classification and intracranial hemorrhage identification. Additionally, the non-parametric regularization module consistently improves the performance of existing state-of-the-art FL approaches. We believe that NPR is a valuable tool in FL under clinical settings.