Autism Spectrum Disorder Detection using Attention-Based CNN and ML Classifiers

Autism spectrum disorder (ASD) is a common condition affecting how children interact and communicate. ASD can manifest in various ways, such as difficulty in social interactions, repetitive behaviours, and verbal and non-verbal communication challenges. These difficulties can significantly impact a child’s ability to form relationships, participate in everyday activities, and navigate social situations. Currently, diagnosing ASD relies a lot on understanding behaviours and developmental disorders, as there are no clear physical markers for diagnosis. This study explores how advanced computer techniques, like deep learning and machine learning, can help detect ASD. Utilizing functional magnetic resonance imaging (fMRI) data from the Autism Brain Imaging Data Exchange (ABIDE) dataset, we have developed a technique to distinguish between ASD accurately and typically developed (TD) individuals. We focus on building a program that can understand the patterns in brain scans using convolutional neural networks (CNNs). Additionally, we used methods like multihead attention and squeeze and excitation to extract more features in the data. We then used machine learning (ML) algorithms to classify the extracted features. The outcomes indicate that our proposed model can detect ASD with an accuracy of 98.59%.