Hybrid Inception V3 and Multi-Layer Perceptron Approach with Extreme Gradient Boost Classifier for Autism Spectrum Disorder Classification

Abstract

Autism Spectrum Disorders (ASD) are complex neuropsychiatric brain disorders characterized by social deficits. They are complex neuropsychiatric disorders characterized by social deficits and repetitive behaviors. Deep learning approaches have been applied to the clinical or behavioral identification of ASD. In this study, we address the classification of ASD using the Autism Brain Imaging Data Exchange I (ABIDE I) database and the Configurable Pipeline for the Analysis of Connectomes (CPAC). We integrated a single volume image generator to overcome limitations related to the small number of samples available in fMRI studies. In addition, we adopted the Inception V3 architecture and ten Multi-Layer Perceptron (MLP) architectures for feature extraction, and used Extreme Gradient Boost (XGBoost) for classification, achieving notable accuracy. In particular, hybridization of Inception V3 with different MLP architectures achieved a maximum accuracy of 96.91% at the Stanford site with a simple MLP architecture. The combination of all MLP architectures with Inception V3 also showed impressive efficiency, achieving an accuracy of up to 92.43%. These models demonstrate an exceptional ability to capture the subtle nuances of resting-state functional MRI (rs-fMRI) data, which is essential for the accurate diagnosis of autism.