Scholarlink

Objective <p>To develop and internally validate machine learning models based on preoperative magnetic resonance imaging (MRI) features for predicting MVD treatment outcomes in TN patients.</p> Methods <p>This retrospective single-center study included 100 patients with TN who underwent primary MVD surgery between January 2020 and December 2024. Preoperative high-resolution MRI sequences including 3D-T2 and time-of-flight magnetic resonance angiography (TOF-MRA) were analyzed. Imaging features included neurovascular compression (NVC) characteristics, nerve morphology, anatomical measurements, and radiomics features. Multiple machine learning algorithms including logistic regression, support vector machine, random forest, and gradient boosting were developed and internally validated using 10-fold cross-validation. All feature preprocessing and selection steps (correlation pruning, univariate analysis, recursive feature elimination) were performed <i>within each training fold</i> of the cross-validation framework to eliminate information leakage. The effective number of predictors after selection and event per predictor (EPP) ratio were calculated to assess overfitting risk associated with class imbalance.</p> Results <p>Complete pain relief was achieved in 93% of patients at final follow-up. The gradient boosting model demonstrated the best predictive performance with an area under the curve (AUC) of 0.847 (95% CI: 0.752–0.925), accuracy of 82.3%, sensitivity of 78.6%, and specificity of 87.5%. Calibration analysis for the gradient boosting model demonstrated a calibration intercept of 0.08 and calibration slope of 0.94, with a Brier score of 0.127 (95% CI: 0.089–0.176), indicating excellent agreement between predicted and observed outcomes. The most important predictive features included NVC severity grade (importance: 0.234), neurovascular angle (importance: 0.187), nerve atrophy presence (importance: 0.156), arterial compression type (importance: 0.142), and proximal NVC location (importance: 0.128). This univariate significance of NVC severity grade (<i>p</i> = 0.003) was consistent with its top ranking in SHAP feature importance analysis, validating the robustness of this key predictive feature. Patients with severe NVC (grade 2) had significantly higher success rates compared to mild NVC (96.4% vs. 76.2%, <i>p</i> = 0.003).</p> Conclusion <p>Machine learning models integrating preoperative MRI features can accurately predict MVD surgical outcomes in TN patients. SHapley Additive exPlanations (SHAP) values provide minimal interpretability for these models, offering basic insight into feature contribution without establishing causal relationships or actionable clinical causality. These predictive models may facilitate personalized treatment planning, improve patient counseling, and optimize surgical decision-making within the evolving framework of precision neurosurgical care and Enhanced Recovery After Surgery (ERAS) protocols. Prospective external validation in multi-center cohorts is urgently required to confirm model generalizability before routine clinical implementation.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Retrospective study: preoperative MRI features for predicting microvascular decompression outcomes in trigeminal neuralgia

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年8月3日

查看全文
  • Fei Chen,
  • Dandan Shang,
  • Dengbin Sun,
  • Zhaoping Wang

摘要

Objective

To develop and internally validate machine learning models based on preoperative magnetic resonance imaging (MRI) features for predicting MVD treatment outcomes in TN patients.

Methods

This retrospective single-center study included 100 patients with TN who underwent primary MVD surgery between January 2020 and December 2024. Preoperative high-resolution MRI sequences including 3D-T2 and time-of-flight magnetic resonance angiography (TOF-MRA) were analyzed. Imaging features included neurovascular compression (NVC) characteristics, nerve morphology, anatomical measurements, and radiomics features. Multiple machine learning algorithms including logistic regression, support vector machine, random forest, and gradient boosting were developed and internally validated using 10-fold cross-validation. All feature preprocessing and selection steps (correlation pruning, univariate analysis, recursive feature elimination) were performed within each training fold of the cross-validation framework to eliminate information leakage. The effective number of predictors after selection and event per predictor (EPP) ratio were calculated to assess overfitting risk associated with class imbalance.

Results

Complete pain relief was achieved in 93% of patients at final follow-up. The gradient boosting model demonstrated the best predictive performance with an area under the curve (AUC) of 0.847 (95% CI: 0.752–0.925), accuracy of 82.3%, sensitivity of 78.6%, and specificity of 87.5%. Calibration analysis for the gradient boosting model demonstrated a calibration intercept of 0.08 and calibration slope of 0.94, with a Brier score of 0.127 (95% CI: 0.089–0.176), indicating excellent agreement between predicted and observed outcomes. The most important predictive features included NVC severity grade (importance: 0.234), neurovascular angle (importance: 0.187), nerve atrophy presence (importance: 0.156), arterial compression type (importance: 0.142), and proximal NVC location (importance: 0.128). This univariate significance of NVC severity grade (p = 0.003) was consistent with its top ranking in SHAP feature importance analysis, validating the robustness of this key predictive feature. Patients with severe NVC (grade 2) had significantly higher success rates compared to mild NVC (96.4% vs. 76.2%, p = 0.003).

Conclusion

Machine learning models integrating preoperative MRI features can accurately predict MVD surgical outcomes in TN patients. SHapley Additive exPlanations (SHAP) values provide minimal interpretability for these models, offering basic insight into feature contribution without establishing causal relationships or actionable clinical causality. These predictive models may facilitate personalized treatment planning, improve patient counseling, and optimize surgical decision-making within the evolving framework of precision neurosurgical care and Enhanced Recovery After Surgery (ERAS) protocols. Prospective external validation in multi-center cohorts is urgently required to confirm model generalizability before routine clinical implementation.