Identification of essential tremor based on resting-state functional connectivity

Area of Science:

• Neuroscience
• Medical Imaging
• Computational Biology

Background:

• Machine learning (ML) shows promise for individual-level clinical diagnosis.
• Essential tremor (ET) diagnosis relies on clinical assessment, lacking objective biomarkers.
• Resting-state functional connectivity (RSFC) measures brain network activity.

Purpose of the Study:

• To determine if whole-brain RSFC metrics and ML can differentiate ET patients from healthy controls (HCs).
• To uncover brain network pathogenesis associated with ET.
• To establish potential diagnostic biomarkers for ET.

Main Methods:

• RSFC data from 127 ET patients and 120 HCs were analyzed.
• Mann-Whitney U test and LASSO methods were used for feature selection.
• Four ML algorithms (SVM, GBDT, RF, GNB) were employed for classification.

Main Results:

• ML algorithms achieved high classification performance: SVM (82.8% accuracy), GBDT (79.4%), RF (78.9%), GNB (72.4%).
• Discriminative features were identified in cerebello-thalamo-motor and non-motor circuits.
• RSFC features correlated with tremor frequency and severity.

Conclusions:

• Combining RSFC metrics with ML algorithms provides accurate discrimination between ET patients and HCs.
• This approach aids in understanding the brain network pathogenesis of ET.
• The study highlights potential RSFC-based biomarkers for ET diagnosis.