Related Experiment Video
Updated: Aug 6, 2026
Closed-Loop Neurostimulation for Biomarker-Driven, Personalized Treatment of Major Depressive Disorder
Published on: July 7, 2023
Second-order morphometric similarity networks predict response to transcutaneous auricular vagus nerve stimulation in
Chunchen Liu1, Yu Xiong2, Tianjiao Xu3
1College of Medical Information and Artificial Intelligence, Jinan Central Hospital Affiliated to Shandong First Medical University, Shandong First Medical University and Shandong Academy of Medical Sciences, Jinan, China.
Introduction:
Transcutaneous auricular vagus nerve stimulation (taVNS) is a promising neuromodulation therapy for major depressive disorder (MDD), but reliable predictors of treatment response are lacking. Second-order morphometric similarity networks (MSN-II), which capture higher-order structural covariance patterns, may provide novel predictive biomarkers.
Methods:
A total of 122 antidepressant-free MDD patients from two centers (Site A, n = 92; Site B, n = 30) underwent structural MRI before and after 8 weeks of taVNS. Baseline MSN-II nodal strength within 26 limbic regions was extracted from T1-weighted images. A LASSO logistic regression model was trained in Site A and externally validated in Site B. Treatment response was defined as ≥60% HAMD-17 reduction.
Results:
MSN-II showed significant training performance (AUC = 0.792 ± 0.158; permutation P < 0.001) and generalized to external validation (AUC = 0.856, 95% CI: 0.693-0.978). MSN-I showed comparable performance (external AUC = 0.804). MSN-II also showed higher external validation performance than ReHo, ALFF, subcortical volumes, and baseline HAMD-17. The left orbitofrontal cortex area 13 (L_OFC_A13; β = -0.649) was the strongest predictor among four retained features. Non-responders showed higher baseline MSN-II in L_OFC_A13 (P = 0.021) and significant post-treatment decreases (P < 0.001), whereas responders remained stable (Time × Group interaction, P = 0.005).
Discussion:
MSN-II features from limbic regions provide promising cross-site prediction of taVNS response in MDD, with L_OFC_A13 emerging as a key biomarker. These findings support the potential of MSN-based approaches for individualized neuromodulation treatment planning.
