Classifying and characterizing nicotine use disorder with high accuracy using machine learning and resting‐state fMRI. (27th June 2018)

• Record Type: Journal Article
• Authors: Wetherill, Reagan R.; Rao, Hengyi; Hager, Nathan; Wang, Jieqiong; Franklin, Teresa R.; Fan, Yong
• Is Part Of: Addiction biology. Volume 24:Number 4(2019)
• Page Start: 811
• Page End: 821
• Publication Date: 2018-06-27
• Subjects: biomarkers -- fMRI -- machine learning -- nicotine -- support vector machines; Substance abuse -- Periodicals; Substance abuse -- Physiological aspects -- Periodicals; Substance-Related Disorders -- periodicals; 616.86
• DOI: 10.1111/adb.12644
• Languages: English
• ISSNs: 1355-6215

Abstract

Abstract: Cigarette smoking continues to be a leading cause of preventable morbidity and mortality. Although the majority of smokers report making a quit attempt in the past year, smoking cessation rates remain modest. Thus, developing accurate, data‐driven methods that can classify and characterize the neural features of nicotine use disorder (NUD) would be a powerful clinical tool that could aid in optimizing treatment development and guide treatment modifications. This investigation applied support vector machine‐based classification to resting‐state functional connectivity (rsFC) data from individuals diagnosed with NUD ( n = 108; 63 male) and matched nonsmoking controls ( n = 108; 63 male) and multi‐dimensional scaling to visualize the heterogeneity of NUD in individual smokers based on rsFC measures. Machine‐based learning models identified five resting‐state networks that played a role in distinguishing smokers from controls: the posterior and anterior default mode networks, the sensorimotor network, the salience network and the right executive control network. The classification method constructed classifiers with an average correct classification rate of 88.1 percent and an average area under the curve of 0.93. Compared with controls, individuals with NUD had weaker functional connectivity measures within these networks ( P < 0.05, false discovery rate corrected). Further, multi‐dimensional scaling visualization demonstrated that controls were similar to each other whereas individuals with NUD had less similarity to controls and to other individuals with NUD. Our findings build upon previous literature demonstrating that machine learning‐based approaches to classifying rsFC data offer a valuable technique to understanding network‐level differences in nicotine‐related neurobiology and extend previous findings by improving classification accuracy and demonstrating the heterogeneity in resting‐state networks of individuals with NUD. Abstract : Machine‐based learning models identified five resting‐state networks that played a role in distinguishing smokers from controls. The classification method constructed classifiers with an average correct classification rate of 88.1 percent and an average area under the curve of 0.93. Multi‐dimensional scaling visualization demonstrated that controls were similar to each other, whereas individuals with nicotine use disorder had less similarity to controls and to other individuals with nicotine use disorder.