Preoperative functional connectivity patterns predict tremor relief following MRgFUS thalamotomy in essential tremor: A machine learning investigation.

BACKGROUND

Magnetic resonance-guided focused ultrasound (MRgFUS) thalamotomy made a breakthrough in treating essential tremor (ET), but with variable tremor responses. This study employed support vector machine regression (SVR) to predict tremor response based on preoperative functional connectivity (FC) patterns.

METHODS

Fifty-six ET patients underwent unilateral MRgFUS thalamotomy and resting-state functional MRI (rs-fMRI). The Clinical Rating Scale for Tremor evaluated 12-month post-treatment responses. Two-sample t-tests identified disease-specific FC differences, which were used in SVR to predict responses. Model performance was evaluated using Pearson's correlation coefficient (r), mean squared error (MSE), and validated via permutation and cross-validation. Longitudinal rs-fMRI in 26 patients examined dynamic changes of the connection with the highest predictive contribution.

RESULTS

Patients demonstrated significant improvement in treated hand tremor at 12-month after thalamotomy (p < 0.001), with a mean improvement of 68%. While baseline tremor severity showed a significant negative correlation with treatment improvement (r = -0.37, p = 0.005), it failed to predict individual outcomes in regression models (linear regression: p = 0.98; SVR: p = 0.46). SVR identified a preoperative FC pattern that predicted tremor response (r = 0.38, p = 0.03; MSE = 0.05, p = 0.02). Longitudinal analysis demonstrated the restoration of the connection with the highest predictive contribution, which increased post-treatment (p = 0.005) and correlated with tremor improvement ratio (r = -0.49, p < 0.001).

CONCLUSION

Preoperative FC patterns predict tremor responses to MRgFUS in ET, serving as a potential imaging biomarker for personalized treatment planning.