Magnetic Resonance Imaging (MRI) plays a key role in detecting physiological abnormalities, but neurological disorders such as attention deficit hyperactivity disorder (ADHD) often lack clearly defined lesion areas. This study proposes a topological feature-based anomaly detection framework for ADHD analysis in MRI. The method constructs a 3D topological space through pixel-wise pathology assessment, uses topological data analysis to enhance subtle structural anomalies, and maps the abstract features back to the original image space for anomaly localization. Experiments on the ADHD-200 dataset show stable performance across site-wise and demographic subgroup evaluations, while identifying lesion regions mainly distributed in the prefrontal-striatal-cerebellar circuitry.
More details about this article are available at this link.