A Robust Autism Brain MRI Classification with GLCM Features and Machine Learning

Abstract

A new robust classification method for analyzing magnetic response images is provided here as a result of the significance of the automated and accurate classification of brain magnetic resonance imaging (MRI] images. Feature extraction, feature selection, and classification are the three stages that make up the system that has been suggested. When we want to extract features from brain sMRI and fMRI, we utilize a technique called Gray Level Co-occurrence Matrix (GLCM), and then we apply recursive feature elimination to choose the most meaningful features. The classifier's goal is to classify subjects' brain sMRI and fMRI as typical (normal) or autistic. A classification with a success of 91.40% accuracy for Child Database, 87.69% for Adolescents Database and 84.12% for Adults database is obtained by Light GBM classifier for sMRI. The Random Forest gives maximum accuracy for Child and Adult database and Light Gradient Boosting Machine gives maximum accuracy for Adolescents database of fMRI. Similarly, a classification with a success of 76.4% and 74% accuracy for Child and Adults database is obtained by Random Forest and 74.29% for adolescents’ database with Light GBM for fMRI database. A robust and efficient technique that minimizes the computational complexity for classification between autism and typical (normal) MRI pictures is the result of the suggested method, which, in comparison to many other recent efforts, results in a robust and efficient strategy.