Classification of ADHD Using Ensemble Algorithms with Deep Learning and Hand Crafted Features

Abstract

Attention Deficit Hyperactivity (ADHD) is a common neurodevelopmental disorder that typically appears in early childhood. Methods developed for diagnosing gives different results at different times. This is a major obstacle in the diagnosis of disease. Diagnosis model of ADHD must be unique, objective, and reliable. In this study, comparative evaluations of both manual and deep features for classification of structural magnetic resonance images is presented. For this purpose, datasets of NPIstanbul Neuropsychiatry Hospital and public datasets of ADHD-200 is used. In order to characterize MRI images First Order, Second Order statictical features and the Alexnet architecture is used. Images are classified with the ensemble algorithm. In order to determine classification performance, accuracy, sensitivity, specificity, tp rate, fp rate and F-measure values are taken into consideration. It was observed that the combination of three manually extracted data sets yielded more successful results in characterizing the data.