Diagnosis Support Model for Autism Spectrum Disorder Using Neuroimaging Data and Xception

Autism Spectrum Disorder is a neurodevelopmental disease-causing significant problem in social and behavioral interactions. Diagnosis of Autism is still challenging for the living community because there is no specific blood test or medical test for its detection. Medical personnel has to diagnose it by checking the individual's past social and physical behavior. In this current scenario, using advanced technological methods like deep learning for the detection of autism will help in solving the problem. It will save a lot of individuals in the early stage. Moreover, brain imaging data for this classification will also help acquire more detection features. This study aims to predict autism using brain scans using a transfer learning algorithm that involves a deep Conv net named Xception. The ABIDE I dataset extracts resting-state functional MRI scans for this study. The dataset consists of brain scans for individuals whose ages lie in the range of 7–64 years, which makes the median age of individuals 14.6 years. The transfer learning approach is used for this classification problem, which uses a deep convolution neural network. The xception algorithm is built on the given dataset for the prediction of autism. This study yielded an accuracy of 97.66% on model training whereas achieved an accuracy of 99.39% on the validation of the model. The Xception transfer learning model gives an accuracy score of 67.35% on the testing of the respective model. All results were obtained from neuroimaging data from the resting state. This paper shows that it is feasible to forecast autism at an early stage using advanced computer algorithms. With the help of the study results, we can conclude that the Xception transfer learning model performed better for the diagnosis of autism.