An Efficient Autism Detection Using Structural Magnetic Resonance Imaging Based on Selective Binary Coded Genetic Algorithm

In this work, an efficient machine learning technique for autism diagnosis using structural magnetic resonance imaging (MRI) is proposed. The proposed technique employs the voxel-based morphometry (VBM) approach to extract a set of 989 relevant features from MRI. These features are used to train an efficient extreme learning machine (ELM) classifier to identify autism spectrum disorder (ASD) and healthy controls. The proposed selective binary coded genetic algorithm (sBCGA) found a subset of significant VBM features. The selected subset of features was used to build a final ELM classifier with maximum overall accuracy. The proposed sBCGA uses a selective sample-balanced crossover designed to improve the classification of ASD and healthy controls. The proposed sBCGA has been extensively tested, and the experimental results clearly indicated better accuracy than existing methods.