Self-Supervised Masked Mesh Learning for Unsupervised Anomaly Detection on 3D Cortical Surfaces

TL;DR

The paper tackles unsupervised anomaly detection on 3D cortical surfaces by introducing a self-supervised masked mesh learning framework (MMN) that learns normative cortical variation from large healthy cohorts. It employs a masked prediction pretraining regime on cortical mesh data, incorporating subject phenotypes, and performs ROI-specific reconstruction-based anomaly scoring to detect dementia-related abnormalities without labeled anomalies. The approach demonstrates that cortical thickness, volume, and sulcal features carry robust signals for distinguishing Alzheimer's disease manifestations from healthy aging, with left-hemisphere regions often more sensitive. This framework offers a scalable, surface-centric method for normative-variation-based anomaly detection in brain imaging, with potential applications to rare diseases and a range of neurological conditions beyond Alzheimer's disease.

Abstract

Unsupervised anomaly detection in brain imaging is challenging. In this paper, we propose self-supervised masked mesh learning for unsupervised anomaly detection on 3D cortical surfaces. Our framework leverages the intrinsic geometry of the cortical surface to learn a self-supervised representation that captures the underlying structure of the brain. We introduce a masked mesh convolutional neural network (MMN) that learns to predict masked regions of the cortical surface. By training the MMN on a large dataset of healthy subjects, we learn a representation that captures the normal variation in the cortical surface. We then use this representation to detect anomalies in unseen individuals by calculating anomaly scores based on the reconstruction error of the MMN. We evaluated our framework by training on population-scale dataset UKB and HCP-Aging and testing on two datasets of Alzheimer's disease patients ADNI and OASIS3. Our results show that our framework can detect anomalies in cortical thickness, cortical volume, and cortical sulcus characteristics, which are known to be biomarkers of Alzheimer's disease. Our proposed framework provides a promising approach for unsupervised anomaly detection based on normative variation of cortical features.

Figures (3)

• Figure 1: Our proposed Masked Mesh Net (MMN) framework.
• Figure 2: Data pipeline for unsupervised anomaly detection.
• Figure 3: Unsupervised anomaly detection result in test set.