Beyond the eye: A relational model for early dementia detection using retinal OCTA images

2024-08-09

Medical Image Anal.

AI Summary

To address the clinical bottleneck of invasive and costly diagnostic procedures for early screening of Alzheimer’s disease (AD) and mild cognitive impairment (MCI), this study proposes a non-invasive relational modeling framework based on retinal optical coherence tomography angiography (OCTA) images. We introduce graph-based relational reasoning to OCTA analysis for the first time, constructing vascular topological graphs from OCTA scans. Our method integrates multi-scale vessel segmentation, topological relation encoding, and weakly supervised contrastive learning to capture global microvascular structural dependencies—overcoming CNNs’ inherent limitation in modeling long-range spatial relationships. Evaluated on a multi-center clinical dataset, our model achieves an AUC of 92.3%—a 6.1% improvement over the state-of-the-art—and attains 89.7% sensitivity for predicting dementia progression within two years. These results significantly enhance the feasibility and generalizability of non-invasive, image-based early AD/MCI detection.

Technology Category

Computer Vision: SegmentationMachine Learning: Graph-based Machine LearningReasoning under Uncertainty: Relational Probabilistic Models

Application Category

Graph Algorithms and Modeling for the Web: Algorithms and analysis for heterogeneous, signed, attributed, multi-relational, temporal, higher-order, and annotated Web-related graphsSearch and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAGUser Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

Problem

• Early dementia detection using retinal OCTA images
• Overcoming invasive, costly diagnostic techniques for AD/MCI
• Exploring eye-brain links for novel diagnostic patterns

Innovation

• PolarNet+ uses OCTA for dementia detection
• Multi-view module analyzes images in 3D
• Graph classification transforms detection task