Autism Spectrum Disorder Detection using Enhanced 3D-ResNet50 Algorithm

Autism Spectrum Disorder (ASD) is a neuro-developmental disease that affects behavioural retardation in verbal communications and social interactions. Clinicians employ various ASD detection techniques to identify the condition. However, these traditional methods are time-consuming and suffer from a lack of accuracy. Over the last two decades, Machine Learning (ML) and Deep Learning (DL) algorithms have played a crucial role in the field of biomedical signal and image processing. In this paper, we propose a Machine Learning Framework. This contains two stages. In the first stage, an enhanced 3D-ResNet50 algorithm is proposed. The proposed algorithm is used to extract features from Magnetic Resonance (MR) Images. In the second stage, the extracted features are used to classify the ASD controls using Machine Learning Algorithms. To improve the accuracy of ASD classification, an enhanced 3D-ResNet50 algorithm is integrated with the ML algorithms. The proposed algorithm is used along with the machine learning algorithms like Support Vector Machine (SVM), K-Nearest Neighbours (KNN), Random Forest (RF), and Logistic Regression (LR). The proposed machine learning framework is tested on 1112 Functional Magnetic Resonance Images (fMRI). These images are collected from the Autism Brain Imaging Data Exchange (ABIDE-I) website. The ABIDE-I website provides a collection of 17 datasets from various international biomedical laboratories. The proposed algorithm is tested on the total ABIDE-I website and 17 individual datasets. Our proposed approach achieved 90% overall accuracy and 97% accuracy for the individual NYU dataset alone.