An Effective Autism Spectrum Disorder Detection Framework Using Adaptive Serial Hybrid Network With Self-Calibrated...

Autism spectrum disorder (ASD) is a medical condition commonly categorized as a neurodevelopmental disorder. Patients with ASD are not limited to common symptoms, and it often requires long-term hospital visits and specialists for definitive evaluation. Research has highlighted the importance of early intervention and care for ASD patients, as it can benefit their lifestyle. Automated detection techniques through deep learners have shown great progress in detecting ASD, as they excel in handling highly complex features and functional activity patterns by evaluating magnetic resonance imaging (MRI). However, as the ASD-associated neuroimaging patterns are not explicitly shown in the MRI, several challenges remain in identifying the ASD related features from MRI evaluation. To overcome these challenges, an advanced deep learning framework has been developed for detecting ASD. In this proposed model, MRI scans are collected from accessible datasets and applied to the feature extraction process. Here, feature extraction was carried out with the support of the proposed Self-Calibrated Dense Convolution Vision Transformer (SC-DCVT). This proposed feature extraction model integrates dense connections and self-calibration mechanisms to capture rich, contextual features from the MRI images, improving the representation of complex patterns. Then, the extracted features are fed to the proposed Adaptive Serial Hybrid Autism Spectrum Disorder Detection Network (ASHASDDetN) model, which integrates two components, such as Residual-Temporal Convolutional Network (Res-TCN) with Dense Conditional Random Field (DCRF), to process ASD detection by capturing relevant features and ensuring its consistency. The reliability of the proposed ASHASDDetN model is enhanced by optimizing the parameters of Res-TCN and DCRF with the support of the Improved Position Updating Actor Optimization Algorithm (IPU-AOA) for the ASD detection task. The effectiveness of the proposed model is proven by conducting extensive experimentation among state-of-the-art techniques.