On the Impact of Data Heterogeneity in Federated Learning Environments with Application to Healthcare Networks

TL;DR

The paper tackles data heterogeneity in federated learning for healthcare by formalizing three main skew types in medical tabular data and proposing a Dirichlet-based mechanism to simulate non-IID distributions. It implements an MQTT-based FL design and benchmarks seven representative FL algorithms across label, quantity, and feature skew using a stroke recurrence dataset, revealing that SCAFFOLD and FedDyn often perform best under heterogeneity while others are preferable under resource constraints. The work provides practical guidelines for algorithm selection in healthcare FL, highlighting the need for prior heterogeneity estimation and offering a real-time networking approach as a novel deployment avenue. Overall, the study advances understanding of how FL algorithms cope with realistic medical data non-IIDness and informs deployment choices for privacy-preserving stroke risk prediction.

Abstract

Federated Learning (FL) allows multiple privacy-sensitive applications to leverage their dataset for a global model construction without any disclosure of the information. One of those domains is healthcare, where groups of silos collaborate in order to generate a global predictor with improved accuracy and generalization. However, the inherent challenge lies in the high heterogeneity of medical data, necessitating sophisticated techniques for assessment and compensation. This paper presents a comprehensive exploration of the mathematical formalization and taxonomy of heterogeneity within FL environments, focusing on the intricacies of medical data. In particular, we address the evaluation and comparison of the most popular FL algorithms with respect to their ability to cope with quantity-based, feature and label distribution-based heterogeneity. The goal is to provide a quantitative evaluation of the impact of data heterogeneity in FL systems for healthcare networks as well as a guideline on FL algorithm selection. Our research extends beyond existing studies by benchmarking seven of the most common FL algorithms against the unique challenges posed by medical data use cases. The paper targets the prediction of the risk of stroke recurrence through a set of tabular clinical reports collected by different federated hospital silos: data heterogeneity frequently encountered in this scenario and its impact on FL performance are discussed.

Figures (4)

• Figure 1: General steps of a Federated Learning process.
• Figure 2: An example of IID medical tabular datasets.
• Figure 3: Heterogeneity in medical tabular datasets data: label (a), quantity (b) and feature (c,d) distribution skew examples
• Figure 4: FL algorithms evaluation on stroke dataset: a), b) - Label distribution skew; c), d) - Quantity skew, e), f) - Feature distribution skew for feature Age; g), h) - Feature distribution skew for feature BMI.