Hybrid groupers-moray orangutan optimization algorithm-based dense-resolution network for autism spectrum disorder detection using multimodal data.

Autism Spectrum Disorder (ASD) is a neurodevelopmental disorder involving cognitive and social impairments. Precise diagnosis is vital for effective intervention. However, conventional diagnostic techniques exhibited poor generalization, and high error rates while dealing with subtle patterns in neuroimaging or behavioral datasets. To resolve these concerns, this research introduces the Groupers and Moray Orangutan Optimization Algorithm based Dense-Resolution Network (GMOA-DRNet). The multimodal inputs consist of Magnetic Resonance Imaging (MRI) brain images and autism-related phenotypic data. Initially, input brain images are pre-processed by Geometric Mean filters and Region of Interest (RoI) extraction. Then, Functional connectivity-based pivotal region are identified using GMOA. After that, feature extraction is performed by Tamura features and Gradient Directional Pattern (GDP). Concurrently, autism-related data is normalized based on the hyperbolic tangent function, and relevant features are selected using ANOVA. Moreover, data augmentation is conducted using the Adaptive Synthetic Sampling Approach (ADASYN) technique. The extracted features and augmented data are used for ASD detection by the DRNet, trained using GMOA, which combines Groupers and Moray Eels Optimization (GMEO) and Orangutan Optimization Algorithm (OOA). Moreover, GMOA-DRNet achieved better performance with an accuracy of 92.77%, sensitivity of 92.52%, specificity of 92.77%, and a False Omission Rate (FOR) of 0.080.