Abstract:

Autism Spectrum Disorder (ASD) is a neurodevelopmental disorder characterized by impairments in social communication and interaction, and repetitive behaviours and interests [5]. Early diagnosis and intervention can significantly improve outcomes for individuals with ASD. A public dataset termed the Autism Brain Imaging Data Exchange (ABIDE) contains neuroimaging and clinical information on people who have or have not ASD. In this study, using the ABIDE-1 dataset, we suggest a Convolutional Neural Network (CNN) model with MobileNet architecture for the detection of ASD. Previous research on the detection of ASD using neuroimaging data has used various machine learning algorithms. However, these algorithms have limitations in terms of scalability and performance. CNNs have been shown to outperform other machine learning algorithms in several image classification tasks, and have recently been applied to neuroimaging data for the detection of ASD. The proposed CNN model with MobileNet architecture was trained and tested on the ABIDE-1 dataset, which contains resting-state functional magnetic resonance imaging (rsfMRI) data from 400 individuals with ASD and 405 typically developing (TD) controls taken from 20 different sites. The dataset was preprocessed to remove noise and artifacts, and the rs-fMRI data were transformed into a connectivity matrix using the Power atlas. The proposed model was trained on 90% of the data and tested on the remaining 10%.

Description:

CNN, ABIDE-1, Adam, ASD