A Cautionary Tale of Self-Supervised Learning for Imaging Biomarkers: Alzheimer's Disease Case Study

2026-01-23

AI Summary

This study addresses the underperformance of existing self-supervised learning methods in discovering imaging biomarkers for Alzheimer’s disease (AD), which lag behind conventional handcrafted features—such as FreeSurfer-derived cortical thickness—in supporting disease classification, progression prediction, and amyloid status assessment. To bridge this gap, the authors propose a Residual Noise Contrastive Estimation (R-NCE) framework that uniquely integrates FreeSurfer anatomical features into the self-supervised objective. By maximizing residual information invariant to data augmentation, R-NCE synergistically combines the strengths of traditional morphometric features with deep representational learning. The method consistently outperforms current approaches across AD classification, conversion prediction, and amyloid positivity estimation. Moreover, its derived metric, R-NCE-BAG, exhibits high heritability, shows significant associations with the MAPT and IRAG1 genes, and is enriched in astrocytes and oligodendrocytes, implicating neurodegenerative and cerebrovascular pathological mechanisms.

Technology Category

Machine Learning: Unsupervised & Self-Supervised LearningComputer Vision: Medical and Biological ImagingCognitive Modeling & Cognitive Systems: Neural Spike Coding

Application Category

Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for rankingGraph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphsWeb Mining and Content Analysis: Robustness and generalizability of Web mining methods

Abstract

Discovery of sensitive and biologically grounded biomarkers is essential for early detection and monitoring of Alzheimer's disease (AD). Structural MRI is widely available but typically relies on hand-crafted features such as cortical thickness or volume. We ask whether self-supervised learning (SSL) can uncover more powerful biomarkers from the same data. Existing SSL methods underperform FreeSurfer-derived features in disease classification, conversion prediction, and amyloid status prediction. We introduce Residual Noise Contrastive Estimation (R-NCE), a new SSL framework that integrates auxiliary FreeSurfer features while maximizing additional augmentation-invariant information. R-NCE outperforms traditional features and existing SSL methods across multiple benchmarks, including AD conversion prediction. To assess biological relevance, we derive Brain Age Gap (BAG) measures and perform genome-wide association studies. R-NCE-BAG shows high heritability and associations with MAPT and IRAG1, with enrichment in astrocytes and oligodendrocytes, indicating sensitivity to neurodegenerative and cerebrovascular processes.

Problem

• Alzheimer's disease
• imaging biomarkers
• structural MRI
• biological relevance
• disease prediction

Innovation

• Self-supervised learning
• Imaging biomarkers
• Residual Noise Contrastive Estimation
• Alzheimer's disease
• Brain Age Gap