Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

FHBench: Towards Efficient and Personalized Federated Learning for Multimodal Healthcare

Penghao Wang, Qian Chen, Teng Zhang, Yingwei Zhang, Wang Lu, Yiqiang Chen

TL;DR

This paper addresses the challenge of privacy-preserving federated learning in healthcare amidst non-IID, multimodal data and limited resources. It introduces FHBench, a realistic multimodal benchmark spanning nervous, cardiovascular, respiratory, and pathology tasks with natural and synthetic non-IID data partitions, and EPFL, a personalized FL framework that uses adaptive LoRA and similarity-weighted aggregation to achieve efficient, robust cross-client learning. Empirical results show EPFL surpassing several baselines across multiple modalities, with notable gains in vision tasks and efficient parameter usage, validating FHBench as a practical evaluation platform. The work advances healthcare FL by coupling a domain-specific benchmark with a scalable, efficient personalization method, enabling more reliable and deployable federated solutions in clinical settings.

Abstract

Federated Learning (FL) has emerged as an effective solution for multi-institutional collaborations without sharing patient data, offering a range of methods tailored for diverse applications. However, real-world medical datasets are often multimodal, and computational resources are limited, posing significant challenges for existing FL approaches. Recognizing these limitations, we developed the Federated Healthcare Benchmark(FHBench), a benchmark specifically designed from datasets derived from real-world healthcare applications. FHBench encompasses critical diagnostic tasks across domains such as the nervous, cardiovascular, and respiratory systems and general pathology, providing comprehensive support for multimodal healthcare evaluations and filling a significant gap in existing benchmarks. Building on FHBench, we introduced Efficient Personalized Federated Learning with Adaptive LoRA(EPFL), a personalized FL framework that demonstrates superior efficiency and effectiveness across various healthcare modalities. Our results highlight the robustness of FHBench as a benchmarking tool and the potential of EPFL as an innovative approach to advancing healthcare-focused FL, addressing key limitations of existing methods.

FHBench: Towards Efficient and Personalized Federated Learning for Multimodal Healthcare

TL;DR

This paper addresses the challenge of privacy-preserving federated learning in healthcare amidst non-IID, multimodal data and limited resources. It introduces FHBench, a realistic multimodal benchmark spanning nervous, cardiovascular, respiratory, and pathology tasks with natural and synthetic non-IID data partitions, and EPFL, a personalized FL framework that uses adaptive LoRA and similarity-weighted aggregation to achieve efficient, robust cross-client learning. Empirical results show EPFL surpassing several baselines across multiple modalities, with notable gains in vision tasks and efficient parameter usage, validating FHBench as a practical evaluation platform. The work advances healthcare FL by coupling a domain-specific benchmark with a scalable, efficient personalization method, enabling more reliable and deployable federated solutions in clinical settings.

Abstract

Federated Learning (FL) has emerged as an effective solution for multi-institutional collaborations without sharing patient data, offering a range of methods tailored for diverse applications. However, real-world medical datasets are often multimodal, and computational resources are limited, posing significant challenges for existing FL approaches. Recognizing these limitations, we developed the Federated Healthcare Benchmark(FHBench), a benchmark specifically designed from datasets derived from real-world healthcare applications. FHBench encompasses critical diagnostic tasks across domains such as the nervous, cardiovascular, and respiratory systems and general pathology, providing comprehensive support for multimodal healthcare evaluations and filling a significant gap in existing benchmarks. Building on FHBench, we introduced Efficient Personalized Federated Learning with Adaptive LoRA(EPFL), a personalized FL framework that demonstrates superior efficiency and effectiveness across various healthcare modalities. Our results highlight the robustness of FHBench as a benchmarking tool and the potential of EPFL as an innovative approach to advancing healthcare-focused FL, addressing key limitations of existing methods.

Paper Structure

This paper contains 19 sections, 9 equations, 6 figures, 11 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Work
  3. FL in Healthcare
  4. Efficicent medical FL with Pre-trained Models
  5. Medicial FL benchmark
  6. FHBench
  7. Objectives
  8. Components
  9. EPFL
  10. Experiments
  11. Results
  12. Ablation study
  13. parameter sensitivity
  14. Efficiency Analysis
  15. Conclusions
  16. ...and 4 more sections

Figures (6)

  • Figure 1: (a) Components of the FHBench framework.(b) Dataset composition in FHBench across multiple modalities.
  • Figure 2: Visualization of Non-IID Data Partitions Across Clients at Different Levels of Heterogeneity, showing the Number of Samples per Class Allocated to Each Client (Indicated by Dot Sizes)
  • Figure 3: Comparison of Full Fine-tuning (left) and LoRA Fine-tuning (right)
  • Figure 4: Impact of Similarity-Weighted Aggregation and Parameter Selection
  • Figure 5: Convergence of Average Accuracy Over Training Rounds
  • ...and 1 more figures