Boundary Corrected Multi-Scale Fusion Network For Real-Time Semantic Segmentation

Autism spectrum disorder (ASD) is a neuro-developmental disorder that results in behavioural retardation in verbal communications and social interactions. Traditional ASD detection methods involve assessing patients' behavioural patterns by medical practitioners, which often lack credibility and precision. The contribution of the current study involves a 3D-CNN (convolutional neural network) model to diagnose ASD patients from healthy individuals using functional magnetic resonance imaging (fMRI) of the brain. We utilised a publicly available dataset, Autism Brain Imaging Data Exchange (ABIDE I), and tested different CNN-based models in individual and combined brain parcellations. Our model showed a better outcome (74.53% accuracy, 69.98% sensitivity, and 76.00% specificity) for combined parcellations than individuals. Further, we compared our model with several state-of-the-art models and discussed the suitability of our model for future prospects. The current model would be a predecessor of future prognosis models or behavioural patterns-based multi-modal models for early detection of ASD.