A classification framework for Autism Spectrum Disorder detection using sMRI: Optimizer based ensemble of deep convolution neural network with on-the-fly data augmentation

Autism Spectrum Disorder (ASD) has affected many children's life due to their hidden symptoms. The late detection of ASD is due to its complex and heterogeneous nature. Due to the noninvasive property and soft tissue information, Magnetic resonance imaging (MRI) has played a very important role in the detection of various brain disorders including ASD. In the past few years, there has been an expeditious increase in the utilization of the Deep Learning approaches in the field of medicine. Many state-of-the-art approaches have utilized Functional Magnetic Resonance Imaging (fMRI) for the detection of ASD, whereas, comparatively very few works have considered Structural Magnetic Resonance Imaging (sMRI) for the detection of ASD with deep learning ap-proaches. This work presents the sMRI-based classification framework for the detection of ASD using optimizer based ensemble of Deep Convolution Neural Network (DCNN) with an on-the-fly data augmentation approach. This work proposes newness toward the approach of ensembling the same model by combining the same DCNN model with different optimizers. The numbers of subjects considered for this work are 484 ASD and 491 Controls. The proposed ensemble model of DCNN with Adam and Nadam optimizer has achieved the accuracy of 77.58%, 77.66%, and 81.35% on the data division ratio of 70:30, 80:20, and 90:10 respectively. Experimental results validate the superior performance of the proposed model compared to the sMRI-based state-of-the-art approaches for the detection of ASD.