Personalized Federated Dictionary Learning for Modeling Heterogeneity in Multi-site fMRI Data

2025-09-24

AI Summary

Multi-center fMRI data exhibit pronounced non-IID characteristics due to privacy constraints and site-specific heterogeneity, severely undermining model generalizability. To address this, we propose Personalized Federated Dictionary Learning (PF-DL): it decomposes each site’s dictionary into shared global atoms and site-specific local atoms. Global atoms are collaboratively updated via federated aggregation to enhance cross-site consistency, while local atoms preserve site-specific neurobiological variability. PF-DL operates without raw data sharing, ensuring privacy-preserving distributed learning and enabling interpretable feature extraction. Experiments on the ABIDE multi-center dataset demonstrate that PF-DL improves classification accuracy by 3.2–5.7% under non-IID conditions and significantly outperforms existing federated learning and dictionary learning methods in robustness. This work establishes a novel paradigm for privacy-aware, interpretable multi-center neuroimaging analysis.

Technology Category

• Machine Learning: Distributed Machine Learning & Federated Learning
• Natural Language Processing: Language Grounding & Multi-modal NLP
• Computer Vision: Multi-modal Vision

Application Category

• User Modeling, Personalization and Recommendation: Federated recommendation systems and personalization
• Responsible Web: Data and user privacy-enhancing technologies for the Web
• Web Mining and Content Analysis: Large pretrained models with web data

Abstract

Data privacy constraints pose significant challenges for large-scale neuroimaging analysis, especially in multi-site functional magnetic resonance imaging (fMRI) studies, where site-specific heterogeneity leads to non-independent and identically distributed (non-IID) data. These factors hinder the development of generalizable models. To address these challenges, we propose Personalized Federated Dictionary Learning (PFedDL), a novel federated learning framework that enables collaborative modeling across sites without sharing raw data. PFedDL performs independent dictionary learning at each site, decomposing each site-specific dictionary into a shared global component and a personalized local component. The global atoms are updated via federated aggregation to promote cross-site consistency, while the local atoms are refined independently to capture site-specific variability, thereby enhancing downstream analysis. Experiments on the ABIDE dataset demonstrate that PFedDL outperforms existing methods in accuracy and robustness across non-IID datasets.

Problem

• Addressing data privacy constraints in multi-site fMRI studies
• Modeling site-specific heterogeneity in non-IID neuroimaging data
• Developing generalizable models without sharing raw data

Innovation

• Federated dictionary learning without sharing raw data
• Decomposing dictionaries into shared global and personalized local components
• Updating global atoms via federated aggregation for cross-site consistency