Continual Learning for fMRI-Based Brain Disorder Diagnosis via Functional Connectivity Matrices Generative Replay

2026-04-15 · Source: Machine Learning · Field: Health & Wellbeing — Medical Specialties & Subspecialties, Medical Devices & Health Technology, Mental Health & Psychological Support · Depth: Expert, quick

Summary

A novel continual learning framework has been developed for fMRI-based brain disorder diagnosis, addressing the challenge of sequential data arrival from heterogeneous clinical sites. This framework introduces a structure-aware variational autoencoder capable of synthesizing realistic functional connectivity (FC) matrices for both patient and control groups. Utilizing this generative model, a multi-level knowledge distillation strategy aligns predictions and graph representations between new-site data and replayed samples. To optimize efficiency, a hierarchical contextual bandit scheme is integrated for adaptive replay sampling. Experimental results on multi-site datasets for major depressive disorder (MDD), schizophrenia (SZ), and autism spectrum disorder (ASD) demonstrate that the generative model improves data augmentation, and the continual learning framework significantly reduces catastrophic forgetting compared to existing methods.

Key takeaway

For AI Scientists developing diagnostic models with fMRI data, this framework offers a robust solution for integrating new clinical site data without retraining from scratch. Your models can maintain performance and avoid catastrophic forgetting by leveraging generative replay and knowledge distillation. Consider implementing a structure-aware VAE and adaptive sampling to enhance data augmentation and learning efficiency in your continual learning pipelines.

Key insights

A continual learning framework uses generative replay and knowledge distillation to diagnose brain disorders from sequential fMRI data.

Principles

• Generative replay mitigates catastrophic forgetting.
• Knowledge distillation aligns new and replayed data.
• Adaptive sampling enhances learning efficiency.

Method

The framework employs a structure-aware VAE for FC matrix synthesis, followed by multi-level knowledge distillation and a hierarchical contextual bandit for adaptive replay sampling.

In practice

• Synthesize FC matrices for data augmentation.
• Apply knowledge distillation for model alignment.
• Use adaptive replay for sequential data.

Topics

• Continual Learning
• fMRI-Based Diagnosis
• Functional Connectivity Matrices
• Generative Replay
• Variational Autoencoder

Code references

• 4me808/FORGE

Best for: AI Scientist, Research Scientist

Related on AIssential

• Treatment-Conditioned Diffusion for Forecasting Neurodegenerative Disease Progression — A treatment-conditioned diffusion framework accurately forecasts neurodegenerative disease progression by integrating imaging and pharmacological data, preserving anatomical detail.
• MOSAIC: Modality-Specific Adaptation for Incremental Continual Learning in Parkinson's Disease Gait Assessment — MOSAIC enables robust continual learning for multimodal Parkinson's gait assessment by addressing modality-specific challenges.
• RECTOR: Masked Region-Channel-Temporal Modeling for Affective and Cognitive Representation Learning — RECTOR unifies region-channel-temporal EEG/sEEG representation learning via self-supervision, adapting functional regions for robust affective and cognitive disorder diagnosis.
• Continual Learning in Modern Hopfield Networks with an Application to Diffusion Models — High-energy, outlier-like samples are more prone to forgetting in continual learning for generative models, and replay is most effective for them.
• Teaching AI to Remember: What I Learned Building a Continual Learning Framework for Drug Discovery — Continual learning models can avoid catastrophic forgetting by integrating new knowledge with awareness of existing representations.
• Distill Once, Adapt Life-Long: Exploring Dataset Distillation for Continual Test-Time Adaptation — DO-ALL uses Dataset Distillation to create stable, privacy-conscious source references for robust continual test-time adaptation.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Machine Learning.