STWO_GDSNet: Supercell Thunderstorm Wind Optimization-Based Gated Dual-Stream Convolutional Neural Network for Autism Spectrum Disorder Detection

Autism spectrum disorder (ASD) affects social interaction and behaviour in children. Current diagnostic methods largely depend on clinical observation and standardized assessments, which can be subjective, time-intensive and influenced by the experience of evaluator. Here, STWO_GDSNet is devised for detecting ASD. Firstly, input image is sent to image pre-preparation, where image is denoised by anisotropic diffusion Kuwahara filtering (ADKF), followed by N4 bias field correction. Next, based on functional connectivity, the pivotal region is extracted by STWO that combines supercell thunderstorm algorithm (STA) and leaf in wind optimization (LiWO). After that, multiscale triangle descriptor (MTD) is extracted in descriptor computation phase. Subsequently, the input autism data are pre-prepared by normalization and standardization. Next, analysis of variance (ANOVA) method conducts attribute pruning, and extracted features as well as the pruned data are passed to gated dual-stream convolutional neural network (GDSNet) for detecting ASD. Here, GDSNet is trained by STWO that incorporates STA and LiWO. The efficacy of STWO_GDSNet is examined by considering accuracy, sensitivity and specificity, which yield maximal values of 96.24%, 95.99% and 96.61%.