Artificial gannet optimization enabled deep convolutional neural network for autism spectrum disorders classification using MRI image

• Published in: Multimedia tools and applications 2024, Vol.83 (30), p.74757-74783
• Main Authors: Prasad, Vadamodula, Ganeshan, R., Rajeswari, R.
• Format: Article
• Language: English
• Subjects: Algorithms; Artificial neural networks; Autism; Classification; Computer Communication Networks; Computer Science; Data Structures and Information Theory; Feature extraction; Image filters; Magnetic resonance imaging; Medical imaging; Multimedia Information Systems; Neural networks; Optimization; Special Purpose and Application-Based Systems
• ISSN: 1573-7721; 1380-7501; 1573-7721
• DOI: 10.1007/s11042-024-19165-0

Autism Spectrum Disorder (ASD) is a neurodevelopment-based disability caused by variations in the brain. This may cause impact on social skills and communication of an individual. Autism is a highly challenging issue to diagnose at the early stages. ASD is one of the important problems to diagnose because it starts manifesting at low ages. This paper aims at ASD classification using MRI Images by Artificial Gannet Optimization enabled Deep Convolutional Neural Network (AGO_DCNN). Here, an MRI image is assumed that be subjected to pre-processing of the image and feature extraction phases as an input. The bilateral filter is employed to remove the noises from the input MRI image and in the pre-processing of an image stage Region of Interest (ROI) extraction is conducted. The pivotal region extraction phase receives a filtered image after which AGO is used to extract the pivotal region. AGO is a newly designed approach by merging two optimizations namely Artificial Ecosystem-based Optimization (AEO) and Gannet Optimization Algorithm (GOA). At the same time, features are extracted from the input image during the feature extraction stage. In a feature extraction stage, some features like texture and statistical features are extracted. At last, the classification of ASD is conducted using DCNN, wherein by AGO the classifier is tuned. Furthermore, AGO_DCNN attained better outcomes in terms of maximal True Positive Rate (TPR), True Negative Rate (TNR), and accuracy values of about 93.7%, 90.8%, 96.8% and False Positive Rate (FPR) and minimal False Negative Rate (FNR) and values of 78.7% and 74.7%.