Multivariate classification of smokers and nonsmokers using SVM‐RFE on structural MRI images

Abstract

Voxel-based morphometry (VBM) studies have revealed gray matter alterations in smokers but this type of analysis has poor predictive value for individual cases, which limits its applicability in clinical diagnoses and treatment. A predictive model would essentially embody a complex biomarker that could be used to evaluate treatment efficacy. In this study, we applied VBM along with a multivariate classification method consisting of a support vector machine with recursive feature elimination to discriminate smokers from nonsmokers using their structural MRI data. Mean gray matter volumes in 1024 cerebral cortical regions of interest created using a sub-parcellated version of the Automated Anatomical Labeling template were calculated from 60 smokers and 60 nonsmokers, and served as input features to the classification procedure. The classifier achieved the highest accuracy of 69.6% when taking the 139 highest ranked features via 10-fold cross-validation. Critically, these features were later validated on an independent testing set that consisted of 28 smokers and 28 nonsmokers, yielding a 64.04% accuracy level (binomial p = 0.01). Following classification, exploratory post-hoc regression analyses were performed, which revealed that gray matter volumes in the putamen, hippocampus, prefrontal cortex, cingulate cortex, caudate, thalamus, pre-/postcentral gyrus, precuneus, and the parahippocampal gyrus, were inversely related to smoking behavioral characteristics. These results not only indicate that smoking related gray matter alterations can provide predictive power for group membership, but also suggest that machine learning techniques can reveal underlying smoking-related neurobiology.

Discriminative analysis of schizophrenia using support vector machine and recursive feature elimination on structural MRI images

“…The machine learning method of SVM has been used to obtain available biomarkers in the diagnoses of psychiatric diseases and addictions. [23,62–64] However, some input features are irrelevant or redundant for classification when performing classification with SVM. Thus, the exclusion of uninformative features from the dataset while retaining discriminative features can not only increase the computation speed but also improve the classification performance.…”

“…[29] A more recent study applied VBM with a multivariate classification method consisting of a SVM with RFE to discriminate smokers from nonsmokers using their structural MRI data and achieved the highest accuracy of 69.6%. [35] …”

“…Our results verified that multiple structural characteristics were useful in increasing the classification performance (see supplementary Table S1). Moreover, in the previous study, [35] the SVM-RFE procedure was applied separately on raw GMVs, mean GMVs from the standard AAL atlas (including 90 cerebral areas) and from the high resolution AAL atlas (including 1024 cerebral areas), and it achieved the highest accuracy when the high-resolution AAL atlas was used. However, our results were inconsistent with this finding and indicated that the classifiers using the input features from the VBM analysis achieved better performances than those using the input features from the ROI analysis.…”

“…Recently, machine learning, which is a type of multivariate analysis that can automatically discriminate individuals within a sample group, was used to classify several neuropsychiatric disorders and addictions, such as SZ, [19–24] Alzheimer's disease , [25–28] depression, [29–31] attention-deficit hyperactivity disorder, [32–34] and smoking. [35] …”

“…[27,39] A recursive feature elimination (RFE) algorithm is an iterative procedure that eliminates 1 backward feature at a time, [40] and thus, can prevent information loss that can occur by eliminating several features in 1 iteration. [41] Previous studies have indicated that RFE employed for dimensionality reduction can significantly improve classification accuracy of neuroimaging data [35] and yield a better classification performance than many other feature reduction methods. [40] …”