Comparative Analysis of Anatomic Plane MRI and Classifier Performance for Autism Spectrum Disorder Classification Using LBP-FOS Features

—This comparative analysis presents a novel approach for Autism Spectrum Disorde (ASD) classification using Local Binary Pattern (LBP) features and First Order Statistics (FOS) features from Magnetic Resonance Imaging (MRI) data. Our study, utilizing the Autism Brain Imaging Data Exchange I (ABIDE I) dataset, implements a comprehensive framework that evaluates multi-plane versus single-plane approaches. Our study implements a comprehensive analysis framework incorporating multiplane (axial, coronal, and sagittal) and single-plane approaches, evaluating four distinct automatic slice selection methods. We utilized distinct machine learning classifiers to determine optimal classification performance across different anatomic planes. Results demonstrate that area-based slice selection with LBP features significantly enhances classification performance, with the multiplane approach consistently outperforming single-plane configurations. Notably, the axial plane yields superior discriminative features compared to other planes. When augmented by a voting mechanism across multiple planes, the Random Forest classifier achieved an accuracy of 92.44%, establishing itself as the most reliable algorithm for ASD classification among the investigated classifiers.