Self-Supervised Cross-Encoder for Neurodegenerative Disease Diagnosis

2025-09-09

AI Summary

Neurodegenerative disease diagnosis from MRI faces two key bottlenecks: heavy reliance on large-scale annotated data and poor interpretability of learned representations. To address these, we propose a self-supervised cross-encoder framework that leverages temporal continuity in longitudinal MRI scans to generate weak supervision signals. Our approach innovatively decouples feature representation into static and dynamic components: static features—capturing stable anatomical structures—are learned via contrastive learning, while dynamic features—modeling pathological progression trajectories—are extracted through input-gradient regularization. This design jointly enhances discriminative power and clinical interpretability. Evaluated on ADNI, our method achieves state-of-the-art classification accuracy. Moreover, it demonstrates superior zero-shot transferability on OASIS and strong cross-task generalization on PPMI, significantly improving model robustness and clinical trustworthiness.

Technology Category

• Machine Learning: Unsupervised & Self-Supervised Learning
• Computer Vision: Representation Learning for Vision
• Natural Language Processing: Language Grounding & Multi-modal NLP

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Responsible Web: Machine-in-the-loop, human agency and autonomy

Abstract

Deep learning has shown significant potential in diagnosing neurodegenerative diseases from MRI data. However, most existing methods rely heavily on large volumes of labeled data and often yield representations that lack interpretability. To address both challenges, we propose a novel self-supervised cross-encoder framework that leverages the temporal continuity in longitudinal MRI scans for supervision. This framework disentangles learned representations into two components: a static representation, constrained by contrastive learning, which captures stable anatomical features; and a dynamic representation, guided by input-gradient regularization, which reflects temporal changes and can be effectively fine-tuned for downstream classification tasks. Experimental results on the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset demonstrate that our method achieves superior classification accuracy and improved interpretability. Furthermore, the learned representations exhibit strong zero-shot generalization on the Open Access Series of Imaging Studies (OASIS) dataset and cross-task generalization on the Parkinson Progression Marker Initiative (PPMI) dataset. The code for the proposed method will be made publicly available.

Problem

• Diagnosing neurodegenerative diseases using MRI data with limited labeled data
• Improving interpretability of learned representations from medical imaging
• Developing models that generalize across different datasets and disease tasks

Innovation

• Self-supervised cross-encoder leveraging temporal MRI continuity
• Disentangles static and dynamic representations via contrastive learning
• Uses input-gradient regularization for interpretable temporal changes