Combined multivariate pattern analysis with frequency-dependent intrinsic brain activity to identify essential tremor

Essential tremor (ET) is the most common tremor disorder, and the intrinsic brain activity changes and diagnostic biomarkers of ET remain unclear. Combined multivariate pattern analysis (MVPA) with resting-state functional MRI (Rs-fMRI) data provides the most promising way to identify individual subjects, reveal brain activity changes, and further establish diagnostic biomarkers in neurological diseases. Using voxel-level amplitude of low-frequency fluctuations (ALFF) and local (regional homogeneity, ReHo) and global (degree centrality, DC) brain connectivity mappings based on three frequency bands (classical band: 0.01-0.10 Hz; slow-5: 0.01-0.023 Hz; slow-4: 0.023-0.073 Hz) of 162 ET patients and 153 well-matched healthy controls (HCs) as input features, MVPA (binary support vector machine, SVM) was performed to differentiate ET from HCs. Each modality achieved good classification performance, except for ReHo based on the slow-4 band with a sensitivity, specificity and total accuracy of 58.64%, 65.36%, 61.90%, respectively (P < 0.05). The classification performance with slow-4 bands was poorer than that with slow-5 and classical bands, but slow-4 bands could be used to reveal the spatial distribution changes in subcortical structures, especially the thalamus. The significant discriminative features were mostly located in the cerebello-thalamo-cortical pathway, and partial correlation analyses showed that significant discriminative features in the cerebello-thalamo-cortical pathway could be used to explain the clinical features of tremor in ET patients. Our findings revealed that voxel-level frequency-dependent ALFF, ReHo and DC could be used to discriminate ET from HCs and help to reveal intrinsic brain activity changes, further acting as potential diagnostic biomarkers.