Scholarlink

<p>To evaluate the diagnostic value of a magnetic resonance imaging (MRI)-based imaging heterogeneity scoring system for differentiating high-grade glioma (HGG) from primary central nervous system lymphoma (PCNSL). This multicenter retrospective study analyzed clinical and preoperative MRI data from 314 pathologically confirmed cases (HGG = 167, PCNSL = 147), comprising 211 patients with single lesions (HGG = 130, PCNSL = 81) and 103 with multifocal lesions (HGG = 37, PCNSL = 66). Patients were randomly assigned to training (single-lesion: n = 147; multifocal: n = 72) and validation (single-lesion: n = 64; multifocal: n = 31) sets in a 7:3 ratio. Distinctive imaging features were used to construct separate logistic regression (LR) models for single-lesion and multifocal-lesion cases, with corresponding scoring systems developed. A baseline model incorporating conventional predictors was developed for comparison. Diagnostic performance was assessed using receiver operating characteristic (ROC) curves (area under the curve [AUC], 95% confidence interval [CI]), Hosmer-Lemeshow tests (goodness-of-fit), calibration curves, and decision curve analysis (DCA). A sensitivity analysis was performed on excluded steroid-treated patients. For single-lesion cases, the training and validation AUCs were 0.940 (95%CI: 0.897–0.983) and 0.908 (0.836–0.981), respectively. Multifocal models achieved training and validation AUCs of 0.960 (0.921–0.999) and 0.927 (0.805–1.000). The heterogeneity scoring system demonstrated significant incremental value over the baseline model (ΔAUC: +0.160–0.290). Hosmer-Lemeshow tests indicated excellent model fit (single-lesion training: χ²= 2.489, <i>P </i>= 0.778; validation: χ² = 6.193, <i>P</i>= 0.185; multifocal training: χ² = 1.760,<i> P </i>= 0.881; validation: χ² = 9.241, <i>P </i>= 0.055). DCA demonstrated substantial net clinical benefit across threshold probabilities. The scoring systems established diagnostic thresholds as follows: ≥ 19 points for HGG (single-lesion) and &gt; 19 points (multifocal), with lower scores indicating PCNSL. Center-stratified validation and repeated cross-validation confirmed strong generalizability across institutions (AUC: 0.934–0.941). The system maintained robust performance in the sensitivity analysis of steroid-treated patients. This MRI heterogeneity-based scoring system provides robust diagnostic accuracy for distinguishing HGG from PCNSL, serving as an objective clinical decision-support tool.</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

Differentiation of high-grade glioma and primary central nervous system lymphoma based on imaging heterogeneity scoring system

  • 期刊论文

  • 发表日期：

    2026年1月23日

查看全文
  • Ming Liu,
  • Jixian Li,
  • Caiqiang Xue,
  • Lei Niu,
  • Song Liu,
  • Yingchao Liu,
  • Shuangshuang Song,
  • Xuejun Liu

摘要

To evaluate the diagnostic value of a magnetic resonance imaging (MRI)-based imaging heterogeneity scoring system for differentiating high-grade glioma (HGG) from primary central nervous system lymphoma (PCNSL). This multicenter retrospective study analyzed clinical and preoperative MRI data from 314 pathologically confirmed cases (HGG = 167, PCNSL = 147), comprising 211 patients with single lesions (HGG = 130, PCNSL = 81) and 103 with multifocal lesions (HGG = 37, PCNSL = 66). Patients were randomly assigned to training (single-lesion: n = 147; multifocal: n = 72) and validation (single-lesion: n = 64; multifocal: n = 31) sets in a 7:3 ratio. Distinctive imaging features were used to construct separate logistic regression (LR) models for single-lesion and multifocal-lesion cases, with corresponding scoring systems developed. A baseline model incorporating conventional predictors was developed for comparison. Diagnostic performance was assessed using receiver operating characteristic (ROC) curves (area under the curve [AUC], 95% confidence interval [CI]), Hosmer-Lemeshow tests (goodness-of-fit), calibration curves, and decision curve analysis (DCA). A sensitivity analysis was performed on excluded steroid-treated patients. For single-lesion cases, the training and validation AUCs were 0.940 (95%CI: 0.897–0.983) and 0.908 (0.836–0.981), respectively. Multifocal models achieved training and validation AUCs of 0.960 (0.921–0.999) and 0.927 (0.805–1.000). The heterogeneity scoring system demonstrated significant incremental value over the baseline model (ΔAUC: +0.160–0.290). Hosmer-Lemeshow tests indicated excellent model fit (single-lesion training: χ²= 2.489, P = 0.778; validation: χ² = 6.193, P= 0.185; multifocal training: χ² = 1.760, P = 0.881; validation: χ² = 9.241, P = 0.055). DCA demonstrated substantial net clinical benefit across threshold probabilities. The scoring systems established diagnostic thresholds as follows: ≥ 19 points for HGG (single-lesion) and > 19 points (multifocal), with lower scores indicating PCNSL. Center-stratified validation and repeated cross-validation confirmed strong generalizability across institutions (AUC: 0.934–0.941). The system maintained robust performance in the sensitivity analysis of steroid-treated patients. This MRI heterogeneity-based scoring system provides robust diagnostic accuracy for distinguishing HGG from PCNSL, serving as an objective clinical decision-support tool.