Machine Learning Approaches for Disease State Classification from Neuroimaging Data

Abstract

Automated recognition and classification of brain diseases are of tremendous value to society. Attention deficit hyperactivity disorder (ADHD) is a diverse spectrum disorder whose clinical diagnosis is based on behavior. In this study, I proposed a two-step cross-validation procedure to illustrate the utility of fully connected cascade (FCC) artificial neural network (ANN) architecture, which provided excellent capability of generalization and outperformed support vector machines in terms of accuracy for both balanced and unbalanced sample sizes, irrespective of the features used. Additionally, I employed various directional and non-directional connectivity based methods to extract discriminative features. I obtained close to 90% accuracy for distinguishing ADHD from healthy subjects and 95% between the ADHD subtypes, which are better than the winning accuracy of the ADHD-200 Global Competition and those reported subsequently. Finally, the most discriminative connectivity features showed reduced and altered connectivity involving the left orbitofrontal cortex and various cerebellar regions in ADHD.