Mental disorders are a major public health concern, affecting millions of people worldwide. The quest for reliable biomarkers for diagnosis and treatment remains a complex challenge. Autism Spectrum Disorder (ASD), a common psychiatric condition, is characterized by atypical patterns in cognitive, emotional, and social domains. Early and accurate diagnosis is crucial to enable effective interventions and improve outcomes for affected individuals.

In this study, we propose an innovative model based on the use of brain atlases (CC200, HO, and AAL) and feature selection, utilizing resting-state functional MRI data from the ABIDE I dataset. Our approach involves feature selection methods to the vectors obtained from flattened connectivity matrices derived from the atlases. These vectors are then used as input for a classifier.

Through cross-validation, we demonstrate that our model outperforms the most advanced methods available for this dataset, achieving an average accuracy of 99.66% in 10-fold cross-validation with the CC200 atlas, using the RFE method and the logistic regression classifier. We also perform feature selection using Lasso and information gain analysis for comparison.