Classification of Cocaine Dependents from fMRI Data Using Cluster-Based Stratification and Deep Learning

Author(s):Show less -

Santos, Jeferson S. ; Savii, Ricardo M. ; Ide, Jaime S. ; Li, Chiang-Shan R. ; Quiles, Marcos G. ; Basgalupp, Marcio P. ; Gervasi, O ; Murgante, B ; Misra, S ; Borruso, G ; Torre, CM ; Rocha, AMAC ; Taniar, D ; Apduhan, BO ; Stankova, E ; Cuzzocrea, A

• Document type: Journal article
• Source: COMPUTATIONAL SCIENCE AND ITS APPLICATIONS - ICCSA 2017, PT I; v. 10404, p. 16-pg., 2017-01-01.

Abstract

Cocaine dependence continues to devastate millions of human lives. According to the 2013 National Survey on Drug Use and Health, approximately 1.5 million Americans are currently addicted to cocaine. It is important to understand how cocaine addicts and non-addicted individuals differ in the functional organization of the brain. This work advances the identification of cocaine dependence based on fMRI classification and innovates by employing deep learning methods. Deep learning has proved its utility in machine learning community, mainly in computational vision and voice recognition. Recently, studies have successfully applied it to fMRI data for brain decoding and classification of pathologies, such as schizophrenia and Alzheimer's disease. These fMRI data were relatively large, and the use of deep learning in small data sets still remains a challenge. In this study, we fill this gap by (i) using Deep Belief Networks and Deep Neural Network to classify cocaine dependents from fMRI, and (ii) presenting a novel stratification method for robust training and evaluation of a relatively small data set. Our results show that deep learning outperforms traditional techniques in most cases, and present a great potential for improvement. (AU)

• FAPESP's process: 16/16291-2 - Characterizing time-varying networks: methods and applications
• Grantee: Marcos Gonçalves Quiles
• Support Opportunities: Scholarships abroad - Research
• FAPESP's process: 16/02870-0 - Multi-objective hyper-heuristics for automatic design of multi-test decision tree induction algorithms
• Grantee: Márcio Porto Basgalupp
• Support Opportunities: Regular Research Grants