Performance Evaluation of Machine Learning Techniques Applied to Magnetic Resonance Imaging of Individuals with Autism Spectrum Disorder

Abstract

Autism Spectrum Disorder (ASD) is a neurodevelopmental disorder, affected by persistent deficits in communication and social interaction and by restricted and repetitive patterns of behavior, interests or activities. Its diagnosis is still a challenge due to the diversity between the manifestations of autistic symptoms, requiring interdisciplinary assessments. This work aims to investigate the performance of the application of techniques of extraction of characteristics and machine learning in magnetic resonance imaging (MRI), in the classification of individuals with ASD. In MRI, the techniques of features extraction were applied: histogram, histogram of oriented gradient and local binary pattern. These features were used to compose the input data of the Support Vector Machine and Artificial Neural Network algorithms. The best result shows an accuracy percentage of 89.66 and a false negative rate of 6.89%. The results obtained suggest that magnetic resonance analysis can contribute to the diagnosis of ASD from the advances in studies in the area.