Autism Spectrum Disorder diagnosis using Multichannel Tri integrated Convolutional Neural Network with Attention

Abstract

The purpose of this research was to use deep learning procedures to detect people with autism spectrum disorder (ASD) using just their brain activity as a diagnostic criterion. We looked at ABIDE (Autism Brain Imaging Data Exchange), a global collection of brain scans from people with ASD. ASD is a neurodevelopment illness that is both difficult to treat and progressive in nature. Most current approaches rely on functional magnetic resonance imaging (fMRI) to make ASD diagnoses; however, these methods only in a limited number of situations, resulting in high accuracy but low generality. We set out to create a deep learning model for the automatic detection of ASD in this study. A multichannel Tri integrated Convolutional Neural Network with Attention Network (MTCAN) was developed to record the connections between different types of data using a combination of neural network layers, a method of attention, and feature fusion. The suggested MTCAN model was tested on a total of 809 participants from the repository. By combining three levels of brain functional connectomes with individual characteristic data, our model outperformed and reached state-of-the-art accuracy of 0.98 in ASD classification. To ensure that the proposed multichannel TCNN model is both generalizable and resilient, further k-fold cross validation was accomplished. The suggested model has an average accuracy of 98%. The suggested model has a sensitivity of 97%, an F1-score of 98%, and specificity 98%. The findings suggest that deep learning models may help automate the clinical diagnosis of ASD in the future.