Individual Structural Covariance Network Predicts Long-Term Motor Improvement in Parkinson Disease with Subthalamic Nucleus Deep Brain Stimulation

Abstract

Background and purpose: The efficacy of long-term chronic subthalamic nucleus deep brain stimulation (STN-DBS) in treating Parkinson disease (PD) exhibits substantial variability among individuals. The preoperative identification of suitable deep brain stimulation (DBS) candidates through predictive means becomes crucial. Our study aims to investigate the predictive value of characterizing individualized structural covariance networks for long-term efficacy of DBS, offering patients a precise and cost-effective preoperative screening tool.

Materials and methods: We included 138 patients with PD and 40 healthy controls. We developed individualized structural covariance networks from T1-weighted images utilizing network template perturbation, and computed the networks' topological characteristics. Patients were categorized according to their long-term motor improvement following STN-DBS. Intergroup analyses were conducted on individual network edges and topological indices, alongside correlation analyses with long-term outcomes for the entire patient cohort. Finally, machine learning algorithms were employed for regression and classification to predict post-DBS motor improvement.

Results: Among the patients with PD, 6 edges (left middle frontal and left caudate nucleus, right olfactory and right insula, left superior medial frontal gyrus and right insula, right middle frontal and left paracentral lobule, right middle frontal and cerebellum, left lobule VIIb of the cerebellum and the vermis of the cerebellum) exhibited significant results in intergroup comparisons and correlation analyses. Increased degree centrality and local efficiency of the cerebellum, parahippocampal gyrus, and postcentral gyrus were associated with DBS improvement. A regression model constructed from these 6 edges revealed a significant correlation between predicted and observed changes in the unified PD rating scale (R = 0.671, P < .001) and receiver operating characteristic analysis demonstrated an area under the curve of 0.802, effectively distinguishing between patients with good and moderate improvement post-DBS.

Conclusions: Our findings reveal the link between individual structural covariance network fingerprints in patients with PD and long-term motor outcome following STN-DBS. Additionally, binary and continuous cerebellum-basal ganglia-frontal structural covariance network edges have emerged as potential predictive biomarkers for DBS motor outcome.

FIG 1.

Schematic outline of the study. A, Gray matter volumes were computed by using the CAT12 toolbox, and gray matter volumes were extracted based on the AAL3 atlas for all HC and patients with PD. B, Individual structural covariance network computation process for patients. PCC indicates Pearson correlation coefficient.

FIG 2.

Clinical Improvement: A, Differences in preoperative and long-term postoperative scores for motor and nonmotor scales in patients with DBS improvement rates in the GIG and MIG. Paired t-tests were conducted for data that passed the normality test, while the Wilcoxon signed-rank test was used for data that did not meet the normality assumption. B, Comparison of the improvement rates in motor and nonmotor symptoms between the GIG and MIG. For normally distributed data, a two-sample t test was employed, whereas the Wilcoxon rank-sum test was used for data that did not meet the normality assumption.

FIG 3.

Intergroup differences in edges and prediction, A, Comparison of intergroup differences in edges between the GIG (blue) and MIG (red) patient groups for edges that are correlated with DBS improvement rate and ranked in the top 1%. Age, sex, TIV, and MoCA were included as covariates. After FDR correction, significant intergroup differences are found in all 6 edges. B, Prediction of long-term improvement groups in patients using 6 edge features.

FIG 4.

The relationship between graph theory metrics and long-term treatment outcomes. Red spherical nodes represent brain regions with P-values corrected for false discovery rate below .05. A, Degree centrality of nodes that correlated with long-term motor prognosis. B, Local efficiency of nodes that correlated with long-term motor prognosis.

FIG 5.

Individual structural covariance networks for predicting long-term treatment outcomes. A, Six edges of correlations with DBS improvement rates, with age, sex, and TIV as covariates. B, Weights of the 6 edges in predicting long-term DBS improvement rates by using the XGBoost model. C, Correlation between the actual and predicted values of long-term DBS improvement rates for the 6 edges by using the XGBoost model.