An Children Autism Spectrum Disorder Diagnosis using Convolutional Neural Network and Recurrent Neural Network

In recent times, Autism is neuro-developmental-disorder categorized by discrepancies in social communication, verbal and non-verbal communication accompanied by partial interests as well as monotonous stereotypical behaviours. The traditional models which Assisted diagnosis of childhood autism i.e. Autism-Spectrum-Disorder (ASD) failed in classification framework of the model and fell short of detecting the nuanced heterogeneity of ASD. Therefore, this paper proposes Convolutional Neural Network- Recurrent Neural Network (CNN-RNN) for ASD detection is employed on Autism Brain Imaging Data Exchange (ABIDE) dataset which incorporates Functional Magnetic Resonance Imaging (fMRI) images as well as effectively detected complex ASD and Non-Autism Spectrum Disorder (Non-ASD) images. Primarily, the required and most relevant features are extracted using CNN by max pooling layers convolutional layers, and dense layers. Then, the dimensionality of the extracted features are reduced by using Pearson Correlation Coefficient (PCC). Ultimately, these features are classified using RNN from fMRI scan images into ASD and Non-ASD images. These images helped in quick identification of children with disorder and children without disorder. The proposed CNN-RNN attained better results in respect of accuracy 99.50%, recall 96.21%, precision 95.23% and f1-score 94.67% when compared with existing Weisfeiler Lehman- Deep Graph Convolutional Networks (WL-Deep GCN).