Control or Autism - Classification using Convolutional Neural Networks on Functional MRI.

Autism spectrum disorders (ASDs) because of it\u0027s permanent nature, high prevalence, substantial heterogeneity, and complexity contributes to a redoubtable challenge to the field of neuroscience and psychiatry. Thus in order to minimize the requirement of Large-scale multidisciplinary efforts, there is a dire need for the development of a reliable and efficient model that gives results at par with the ones offered by the doctors based on symptomatology. Many significant works have been propagated for classification of ASD, carried out over the Resting-State functional MRI (RS-fMRI) data. A novel convolutional neural network architecture has been developed for substantially analyzing the similarity in brain neural connectivities of the two classes, i.e. autism and control that outperforms existing Machine Learning/Deep Learning methods and produces state-of-the-art (SOTA) results. We have been able to attain an accuracy of $0.76\\pm 0.039$ , precision of $0.7863\\pm 0.037$ , and specificity of $0.8169\\pm 0.047$ using ten-fold Cross-validation policy on the pre-processed version of RS-fMRI data from the ABIDE-I database.