Multivariate Pattern Classification of Primary Insomnia Using Three Types of Functional Connectivity Features

ObjectiveTo investigate whether or not functional connectivity (FC) could be used as a potential biomarker for classification of primary insomnia (PI) at the individual level by using multivariate pattern analysis (MVPA). MethodsThirty-eight drug-naive patients with PI and 44 healthy controls (HC) underwent resting-state functional MR imaging. Three commonly used FC metrics were calculated for each participant. We used the MVPA framework using linear support vector machine (SVM) with the three types of metrics as features separately. Subsequently, an unbiased N-fold cross-validation strategy was used to generate a classification system and was then used to evaluate its classification performances. Finally, FC metrics with significant high classification performance were compared between the two groups and were correlated with clinical characteristics, i.e., Insomnia Severity Index (ISI), Pittsburgh Sleep Quality Index (PSQI), Self-rating Anxiety Scale (SAS), Self-rating Depression Scale (SDS). ResultsThe best classifier could reach up to an accuracy of 81.5%, with sensitivity of 84.9%, specificity of 79.1% and area under the receiver operating characteristic curve (AUC) of 83.0% (all P < 0.001). Right fronto-insular cortex, left precuneus and left middle frontal gyrus showed high classification weights. In addition, right fronto-insular cortex and left middle frontal gyrus were the overlapping regions between MVPA and group comparison. Correlation analysis showed that functional connectivity strength (FCS) in left middle frontal gyrus and head of right caudate nucleus were correlated with PSQI and SDS respectively. ConclusionThe current study suggests abnormal FCS might serve as a potential neuromarkers for PI. Key PointsFCS in fronto-insular cortex and middle frontal gyrus may be a neuroimaging biomarker for insomnia. FCS can be used to distinguish between patients with primary insomnia from healthy controls with high classification accuracy (81.5%; P < 0.001). FCS in left middle frontal gyrus and head of right caudate nucleus were correlated with PSQI and SDS respectively. Abbreviations