A Novel Image Processing Scheme Design to Predict Autism Spectrum Disorder Using Deep Learning Mechanism

Autism Spectrum Disorder (ASD) is a neurodevelopmental condition characterized by challenges in social communication and the presence of restricted, repetitive behaviors. This paper presents a novel image processing scheme for predicting ASD using a hybrid deep learning mechanism applied to neuroimaging data. The proposed model integrates EfficientNet for feature extraction, an AutoEncoder for dimensionality reduction, ShuffleNet for classification, and Transformer networks for sequence modeling, combining the strengths of these architectures in an ensemble learning framework. The ABIDE dataset, comprising MRI and fMRI scans of ASD and neurotypical individuals, was used for model training and evaluation. To ensure robust performance, we applied advanced preprocessing techniques such as contrast enhancement, segmentation, and noise reduction to the neuroimaging data. The proposed model achieved an accuracy of 96.33%, outperforming all compared models. This method demonstrates potential for enhancing early diagnosis of ASD, facilitating timely interventions and improving patient outcomes.