DynFS-MoE: Dynamic Functional-Structural Mixture-of-Experts for Post-Traumatic Epilepsy Diagnosis

2026-06-15 · Source: Computer Vision and Pattern Recognition · Field: Science & Research — Health & Medical Research, Artificial Intelligence & Machine Learning, Computer Vision & Pattern Recognition · Depth: Expert, quick

Summary

The DynFS-MoE (Dynamic Functional-Structural Mixture-of-Experts) framework is proposed for the early diagnosis of Post-Traumatic Epilepsy (PTE), a severe complication of traumatic brain injury (TBI). This multimodal framework integrates functional and structural MRI data over time, employing time-aware functional-structural encoding and class-conditioned expert routing. It utilizes modality-specific and cross-modal experts to learn complementary representations, while a Modality-Class MoE (MCoE) module dynamically dispatches expert weights based on specific classification objectives. Experimental results across three binary classification tasks demonstrate that DynFS-MoE consistently outperforms static fusion baselines. Furthermore, high-interpretability analyses reveal meaningful region-of-interest (ROI) interactions, providing an interpretable approach for PTE diagnosis and risk stratification by effectively capturing class-dependent brain interaction patterns.

Key takeaway

For AI Scientists and Machine Learning Engineers developing diagnostic models for complex neurological conditions like PTE, you should consider dynamic multimodal expert frameworks. This approach, exemplified by DynFS-MoE, significantly outperforms static fusion methods by capturing class-dependent brain interaction patterns. Implement time-aware functional-structural encoding and class-conditioned expert routing to improve diagnostic accuracy and gain interpretable insights into disease mechanisms, enhancing risk stratification and early intervention strategies.

Key insights

DynFS-MoE dynamically integrates multimodal MRI data via expert routing for interpretable, early post-traumatic epilepsy diagnosis.

Principles

• Dynamic multimodal integration improves diagnostic accuracy.
• Class-conditioned expert routing enhances model specificity.
• Interpretable ROI interactions reveal disease patterns.

Method

DynFS-MoE uses time-aware functional-structural encoding and class-conditioned expert routing to integrate functional and structural MRI, with an MCoE module dynamically dispatching expert weights for classification.

In practice

• Apply dynamic MoE for complex disease diagnosis.
• Use multimodal MRI for enhanced brain analysis.
• Implement class-conditioned routing for specific tasks.

Topics

• Post-Traumatic Epilepsy Diagnosis
• Mixture-of-Experts
• Multimodal MRI
• Traumatic Brain Injury
• Medical Imaging Analysis
• Interpretable AI

Best for: AI Scientist, Machine Learning Engineer, Research Scientist

Related on AIssential

• Predicting Post-Traumatic Epilepsy from Clinical Records using Large Language Model Embeddings — LLM embeddings from clinical records can predict post-traumatic epilepsy with high accuracy.
• Fusion Learning from Dynamic Functional Connectivity: Combining the Amplitude and Phase of fMRI Signals to Identify Brain Disorders — Combining fMRI signal amplitude and phase via fusion learning improves brain disorder detection.
• Seizure-Semiology-Suite (S3): A Clinically Multimodal Dataset, Benchmark, and Models for Seizure Semiology Understanding — MLLMs struggle with fine-grained medical video interpretation, but domain-specific methods significantly boost diagnostic accuracy for seizure semiology.
• BrainDINO: A Brain MRI Foundation Model for Generalizable Clinical Representation Learning — Large-scale slice-wise self-supervised learning yields a unified, transferable brain MRI representation for diverse neuroimaging tasks.
• 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.
• Case-Aware Medical Image Classification with Multimodal Knowledge Graphs and Reliability-Guided Refinement — Medical image classification improves by integrating multimodal knowledge graphs from similar cases with reliability-guided refinement.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Computer Vision and Pattern Recognition.