Deep Feature Selection and Multi-Sparsity Functional Connectivity Networks for Autism Classification

Abstract Autism spectrum disorder (ASD) is associated with heterogeneous alterations in functional brain organization, making reliable neuroimaging-based characterization difficult. We developed a framework combining deep feature selection with multi-sparsity functional connectivity networks to identify discriminative connectivity patterns and classify ASD. Resting-state functional magnetic resonance imaging data from 871 participants in the Autism Brain Imaging Data Exchange I dataset (403 ASD and 468 typically developing controls) were analyzed. Pearson correlation networks were constructed across 111 regions and thresholded at multiple sparsity levels. Deep feature selection identified stable connections across training folds, while nested cross-validation selected the optimal sparsity level and evaluated a linear support vector machine. The framework achieved 86.2\% accuracy, 88.8\% sensitivity, 84.0\% specificity, and an area under the curve of 91.6\%. Discriminative connections were concentrated in frontal, occipital, and parietal regions, including the superior frontal gyrus and lateral occipital cortex. These findings show that combining feature selection with multi-sparsity connectome analysis can improve the identification of reproducible autism-associated connectivity patterns. The results should be interpreted as research-level classification rather than a clinical diagnostic test and require independent external validation.