Identification of essential tremor based on resting-state functional connectivity.

Currently, machine-learning algorithms have been considered the most promising approach to reach a clinical diagnosis at the individual level. This study aimed to investigate whether the whole-brain resting-state functional connectivity (RSFC) metrics combined with machine-learning algorithms could be used to identify essential tremor (ET) patients from healthy controls (HCs) and further revealed ET-related brain network pathogenesis to establish the potential diagnostic biomarkers. The RSFC metrics obtained from 127 ET patients and 120 HCs were used as input features, then the Mann-Whitney U test and the least absolute shrinkage and selection operator (LASSO) methods were applied to reduce feature dimensionality. Four machine-learning algorithms were adopted to identify ET from HCs. The accuracy, sensitivity, specificity and the area under the curve (AUC) were used to evaluate the classification performances. The support vector machine, gradient boosting decision tree, random forest and Gaussian naïve Bayes algorithms could achieve good classification performances with accuracy at 82.8%, 79.4%, 78.9% and 72.4%, respectively. The most discriminative features were primarily located in the cerebello-thalamo-motor and non-motor circuits. Correlation analysis showed that two RSFC features were positively correlated with tremor frequency and four RSFC features were negatively correlated with tremor severity. The present study demonstrated that combining the RSFC matrices with multiple machine-learning algorithms could not only achieve high classification accuracy for discriminating ET patients from HCs but also help us to reveal the potential brain network pathogenesis in ET.