Copyright © 2026 Authors retain the copyright of this article. This article is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

@article{199572,
        author = {Navya Durga and Dr. V. Usha bala and Jayanth vorloth and K. Sandeep and A.Srinivas sai},
        title = {A Two-Layer Approach for Behavioral Assessment and Classification of Neurological Disorders Using Resting-State fMRI},
        journal = {International Journal of Innovative Research in Technology},
        year = {2026},
        volume = {12},
        number = {11},
        pages = {16016-16029},
        issn = {2349-6002},
        url = {https://ijirt.org/article?manuscript=199572},
        abstract = {Brain-related conditions including Autism Spectrum Disorder (ASD), Attention Deficit Hyperactivity Disorder (ADHD), Schizophrenia (SCHZ), and Bipolar Disorder affect an enormous number of people worldwide, yet getting an accurate diagnosis still depends heavily on clinical judgment, behavioral interviews, and subjective rating scales. There are currently no objective biological tests in routine use that can reliably distinguish these conditions from one another. Resting-state functional MRI (rs-fMRI) offers a promising avenue forward, as it records spontaneous brain activity without requiring any active participation from the patient, making it suitable even for individuals with limited cooperation or cognitive capacity [1]. That said, working with rs-fMRI data is technically challenging because the brain does not behave like a regular grid—it is a network, and the patterns that separate one neurological condition from another are encoded in how regions connect with each other, not just in the signals from individual regions considered in isolation. Standard neural network approaches that assume spatial regularity simply do not capture this structure well. To address this, we developed a hierarchical graph-based deep learning system that treats functional connectivity as what it truly is: a weighted network, processed by architectures specifically built for graph-structured data. Our model uses Graph Attention Networks version 2 (GATv2) combined with Self-Attention Graph Pooling (SAGPooling) in a two-stage architecture that progressively focuses on the most diagnostically meaningful brain sub-networks. On top of that, we extract multiple graph snapshots per scan using a sliding time window, which helps capture how connectivity shifts over time rather than just averaging everything into a single static picture. Tested on 500 subjects drawn from the ABIDE and ADHD-200 datasets across five classes—healthy controls, ASD, ADHD, SCHZ, and Bipolar—our framework achieves 83.1% classification accuracy, a weighted F1 of 0.824, and a macro AUC of 0.921. These numbers represent gains of 11.2 percentage points over a standard Graph Convolutional Network and about 20 points over a linear SVM. All improvements are statistically significant, and the brain regions the model prioritizes closely match what decades of neuroscience research has already identified for each condition.},
        keywords = {Graph Attention Networks, fMRI, Brain Connectivity, Neurological Disorder Classification, SAG Pooling, GATv2, Deep Learning, Functional Connectivity, Biomarker Extraction, Dynamic Connectivity.},
        month = {April},
        }