Machine and Deep Learning Based Detection of Attention Deficit Hyperactivity Disorder

Attention Deficit Hyperactivity Disorder (ADHD) is a brain disease that can cause academic, social and psychiatric problems. Structural Magnetic resonance imaging (structural MR) is a critical diagnostic tool used to examine brain anatomy and pathology. In this study, an objective ADHD detection model was developed with Machine Learning (ML) and Deep Learning (DL) methods using structural MR images. Gray and white matter is an important parameter in the diagnosis of many psychiatric diseases. An algorithm has been developed for the detection of slices in which gray and white matter appear complete and clear. While the slices determined by the proposed algorithm are assigned to one dataset, all slices of the structural MR image are assigned to the other dataset. Different feature sets were created by characterizing structural MR images with ML (LBP and Haralick) and DL methods (AlexNet, VggNet, ResNet, SqueezeNet and InceptionResNet). High classification performances were observed in the characterization of the dataset containing the selected slices with ML and DL algorithms. High classification performances were observed with LBP and Haralick, which were especially successful in capturing changes in texture.