A Machine Learning Approach for Autism Spectrum Disorder Detection using BOLD-fMRI Signals and ABIDE-II Dataset

Autism spectrum disorder (ASD) is a neurodevelopmental disorder affecting nearly 1 in 60 children worldwide. The main issue in ASD treatment has no definite cure and the only solution to this is to control the disorder’s progression. Traditional ASD diagnosis mainly relies on subjective measures which is wholly a behavioural-dependent approach and may lead to delayed diagnosis. To overcome this, existing studies have shifted towards neuroimaging biomarkers which are more objective and can help in detecting ASD at an early stage. The implementation of machine learning techniques as a diagnostic tool for ASD is rapidly growing in the field of neuroscience and has greater possibilities in discriminating between ASD and controlled children. Several automated models for effective and early ASD diagnosis using machine learning and fMRI have been reported. However, these models barely reach an accuracy of nearly upto 60-70 percent. Most of these ASD studies have been conducted using ABIDE I database that has been created earlier. In this study, we presented a classification technique to analyze functional brain connectivity by utilizing the newly developed database ABIDE II which is a multisite data aggregated from three different sites. Various classification algorithms: Support Vector Machine (SVM), Logistic Regression (LR) and Random forest (RF) have been employed and the results show that the RF outperforms the other two techniques by yielding an optimal classification accuracy of 75 percent which is much more relative to the previous works.