Autism detection based on MRI images using deep learning

Abstract

Autism spectrum disorder (ASD) is a neurodysfunction or neurodevelopmental disorder. This causes a patient to have trouble with social interaction which causes social instability. It also causes speech problems or difficulty with any sort of verbal communication as well as nonverbal communication. The primary challenge associated with autism lies in its intricacy of early diagnosis. The difficulty in diagnosing is due to the lack of a proper medical test for it. Researchers have yet to discover a biomarker or specific gene that can detect autism. Doctors still use outdated methods to identify autism nowadays. In order to tackle this issue and facilitate the diagnosis of autism, we employed deep learning techniques to create a method for ASD diagnosis. This study presents a deep learning-based approach using Functional MRI and Structural MRI images. We used the ABIDE dataset for this research. After examining the MRI pictures, a method was developed to pick out particular layers from the MRI images. Our dataset was then constructed using images from ABIDE for our models to train and test without performing any pre-processing. Various cutting-edge deep learning architectures were chosen to train using our created dataset. Novel architectures were used to attain an accuracy of 80% to practically 84%. A custom block was used later in the research to expand the dataset and achieve more accuracy. Based on our findings, it is evident that the deep learning model with our modified block demonstrates superior performance when compared to alternative techniques. Consequently, it offers a feasible solution for the diagnosis of ASD using MRI images.