Graph Neural Networks for Neurological Disorder Detection

Background and Aims: Graph Neural Networks (GNNs) have emerged as the game-changing tool to process complex neurologic information by presenting unparalleled models of non-Euclidean patterns of connectivity in brains. This paper introduces the Integrated Functional Connectivity GNN (IFC-GNN), a more advanced architecture that dynamically constructs latent features from several neuroimaging modalities within an integrated framework, adding time dependencies and population demographics to enhance biomarker discoveries. The efficacy of IFC-GNN for neuro-disorder diagnosis was demonstrated through three tasks: (۱) Autism Spectrum Disorder (ASD) classification from Functional Magnetic Resonance Imaging (fMRI) data in the ABIDE dataset, ۸۰.۶۶% accuracy; (۲) Alzheimer's Disease (AD) progression prediction from Electroencephalogram (EEG)-based networks, where graph transformers attained ۹۱.۷۷% accuracy; and (۳) Epilepsy seizure focus localization, ۹۹.۸۴% accuracy with graph transformers. The IFC-GNN outperformed conventional GNNs and methods like CNNs and SVMs, revealing significant biomarkers like reduced alpha-band power in AD and disrupted theta-band synchronization in Epilepsy. By combining spatial-temporal brain connectivity analysis with demographic data, IFC-GNN drives precision medicine, yielding clinically actionable insights into early diagnosis and therapeutic targeting. This paper highlights the promise of GNNs in computational neurology and encourages interpretable, multimodal architectures to help close the loop between deep learning and clinical practice.

کلیدواژه ها:

Graph Neural Networks ، Autism Spectrum ، Alzheimer