Autism spectrum disorder detection using parallel DCNN with improved teaching learning optimization feature selection scheme

The identification of a neurological disorder known as autism spectrum disorder (ASD) is essential and vital for improving the quality of life and providing appropriate medical care for those with autism. Good health and well-being are essential for individuals with autism, just like anyone else. In the last decade, numerous machine learning (ML) and deep learning (DL) based techniques and methods were used for Autism Disorder Detection (ASD) with the help of magnetic resonance images (MRI). The performance of this technique is susceptible to poor feature representation, redundant features, complexity of DL frameworks, and poor visual quality of the images. This paper presents ASDD based on a parallel Deep Convolution Neural Network (PDCNN). It includes image enhancement, feature extraction, feature selection, deep feature representation, and ASDD. It presents an improved double-stage Gaussian Weiner Filtering scheme to minimize blur, contrast, and uneven illumination in some images. Further, it offers the shape and texture feature extraction of functional MRI (fMRI) with gray level co-occurrence matrix (GLCM), local binary pattern (LBP), and histogram of oriented gradient (HOG), and local directional pattern (LDP). Afterward, an improved teaching-learning-based scheme is utilized to select prominent features to minimize the computational intricacy of the PDCNN. The outcomes of the system are validated on the ABIDE-I dataset.