Scholarlink

Purpose <p>This study aimed to develop and validate a predictive model integrating paravertebral muscle radiomic features and clinical parameters to assess the risk of recurrent vertebral compression fractures following percutaneous vertebroplasty (PVP).</p> Materials and methods <p>A retrospective study was conducted on 134 patients with spinal compression fractures who underwent PVP between June 2022 and June 2023. Patients were randomly divided into a training group (<i>n</i> = 93) and a validation group (<i>n</i> = 41), while clinical and imaging data from an additional 32 patients were collected as a testing group. Radiomic features were extracted using Python, followed by dimensionality reduction via the Mann-Whitney U test and feature selection through LASSO regression. A radiomics score (Rad-score) was constructed and integrated with clinical parameters. Univariate and multivariate logistic regression analyses were employed to identify risk factors and establish the predictive model. The model’s performance was evaluated using ROC curves, calibration curves (Calibration curves) and decision curves (DCA).</p> Results <p><?tk 4?>A total of 864 radiomic features were extracted. After stability analysis and filtering, LASSO regression identified eight significant features for constructing the Rad-score. Multivariate analysis guided by AIC identified Rad-score, bone mineral density (BMD), and intervertebral bone cement leakage as independent predictors of recurrent fractures. The combined model incorporating these factors achieved AUC values of 0.888 in the training cohort, 0.843 in the validation cohort, and 0.817 in the test cohort. Calibration curves indicated good model fit, particularly in the training and validation sets. DCA demonstrated the clinical utility of the combined model across a range of probability thresholds.</p> Conclusion <p>A multimodal predictive model integrating radiomics features derived from paravertebral muscles with clinical risk factors (BMD and cement leakage) can effectively stratify the risk of recurrent vertebral fractures following PVP. This approach offers potential improvements in predictive accuracy compared to traditional clinical models alone.</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

Development of a multimodal radiomics-based model for predicting recurrent fracture risk after PVP

  • 期刊论文

  • 发表日期：

    2026年2月16日

查看全文
  • Muhan Liu,
  • Xiaoning Yang,
  • Yingshan Ding,
  • Zheyi Xiao

摘要

Purpose

This study aimed to develop and validate a predictive model integrating paravertebral muscle radiomic features and clinical parameters to assess the risk of recurrent vertebral compression fractures following percutaneous vertebroplasty (PVP).

Materials and methods

A retrospective study was conducted on 134 patients with spinal compression fractures who underwent PVP between June 2022 and June 2023. Patients were randomly divided into a training group (n = 93) and a validation group (n = 41), while clinical and imaging data from an additional 32 patients were collected as a testing group. Radiomic features were extracted using Python, followed by dimensionality reduction via the Mann-Whitney U test and feature selection through LASSO regression. A radiomics score (Rad-score) was constructed and integrated with clinical parameters. Univariate and multivariate logistic regression analyses were employed to identify risk factors and establish the predictive model. The model’s performance was evaluated using ROC curves, calibration curves (Calibration curves) and decision curves (DCA).

Results

A total of 864 radiomic features were extracted. After stability analysis and filtering, LASSO regression identified eight significant features for constructing the Rad-score. Multivariate analysis guided by AIC identified Rad-score, bone mineral density (BMD), and intervertebral bone cement leakage as independent predictors of recurrent fractures. The combined model incorporating these factors achieved AUC values of 0.888 in the training cohort, 0.843 in the validation cohort, and 0.817 in the test cohort. Calibration curves indicated good model fit, particularly in the training and validation sets. DCA demonstrated the clinical utility of the combined model across a range of probability thresholds.

Conclusion

A multimodal predictive model integrating radiomics features derived from paravertebral muscles with clinical risk factors (BMD and cement leakage) can effectively stratify the risk of recurrent vertebral fractures following PVP. This approach offers potential improvements in predictive accuracy compared to traditional clinical models alone.