Scholarlink

<p>Differentiating primary lumbar spinal tumors (benign or malignant) from metastatic lesions before invasive biopsy supports tailored clinical management planning, yet conventional qualitative MRI interpretation frequently yields overlapping imaging manifestations with limited discriminatory capacity. This dual-center retrospective observational study constructed and internally validated a multiparametric MRI combined clinical-radiomics nomogram for binary preoperative risk stratification between primary and metastatic lumbar spinal lesions. A total of 100 pathologically confirmed patients scanned with standardized T1WI, T2WI and FS-T2WI sequences between January 2020 and December 2023 were enrolled. After standardized VOI segmentation and IBSI-compliant feature extraction (327 raw radiomic features per patient), a dual-stage feature filtering pipeline combining mRMR screening and LASSO penalized regression with balanced class weights was implemented within a 5-fold nested cross-validation framework to mitigate overfitting risk. The optimized radiomics signature (Rad-score) incorporating 9 stable imaging features was integrated with independent clinical predictors identified via multivariate regression to generate a visualized binary nomogram. Bootstrap optimism correction (1000 resamples) yielded a corrected validation cohort AUC of 0.872 (95% CI: 0.831–0.908). The integrated model demonstrated acceptable calibration with bootstrap-corrected Brier score = 0.116 and calibration slope = 0.941. Decision curve analysis indicated consistent positive net clinical benefit across threshold probabilities ranging from 0.22 to 0.91. Subgroup analysis revealed moderate performance fluctuation across metastatic lesions originating from distinct primary visceral malignancies. The proposed multiparametric MRI radiomics nomogram provides quantitative non-invasive stratification evidence to complement routine radiological assessment. Further prospective geographically independent multi-cohort validation is required to confirm external generalizability before routine clinical utilization.</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

Multiparametric MRI radiomics nomogram for binary preoperative stratification of primary versus metastatic lumbar spinal tumors

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年8月5日

查看全文
  • Canghai Shen,
  • Shuai Yang,
  • Xi Chen,
  • Yongjian Feng,
  • Yancheng Song,
  • Jianxi Zhou,
  • Yunchuan Sun

摘要

Differentiating primary lumbar spinal tumors (benign or malignant) from metastatic lesions before invasive biopsy supports tailored clinical management planning, yet conventional qualitative MRI interpretation frequently yields overlapping imaging manifestations with limited discriminatory capacity. This dual-center retrospective observational study constructed and internally validated a multiparametric MRI combined clinical-radiomics nomogram for binary preoperative risk stratification between primary and metastatic lumbar spinal lesions. A total of 100 pathologically confirmed patients scanned with standardized T1WI, T2WI and FS-T2WI sequences between January 2020 and December 2023 were enrolled. After standardized VOI segmentation and IBSI-compliant feature extraction (327 raw radiomic features per patient), a dual-stage feature filtering pipeline combining mRMR screening and LASSO penalized regression with balanced class weights was implemented within a 5-fold nested cross-validation framework to mitigate overfitting risk. The optimized radiomics signature (Rad-score) incorporating 9 stable imaging features was integrated with independent clinical predictors identified via multivariate regression to generate a visualized binary nomogram. Bootstrap optimism correction (1000 resamples) yielded a corrected validation cohort AUC of 0.872 (95% CI: 0.831–0.908). The integrated model demonstrated acceptable calibration with bootstrap-corrected Brier score = 0.116 and calibration slope = 0.941. Decision curve analysis indicated consistent positive net clinical benefit across threshold probabilities ranging from 0.22 to 0.91. Subgroup analysis revealed moderate performance fluctuation across metastatic lesions originating from distinct primary visceral malignancies. The proposed multiparametric MRI radiomics nomogram provides quantitative non-invasive stratification evidence to complement routine radiological assessment. Further prospective geographically independent multi-cohort validation is required to confirm external generalizability before routine clinical utilization.