Scholarlink

Background <p>To investigate the feasibility of MRI-based habitat radiomics to differentiate indeterminate Grade 3 nasopharyngeal lesions into nasopharyngeal lymphoid hyperplasia or early-stage nasopharyngeal carcinoma, and to explore its complementary value for radiologists in diagnostically ambiguous cases.</p> Methods <p>In this retrospective study, 138 patients with Grade 3 nasopharyngeal lesions who underwent contrast-enhanced MRI and endoscopic biopsy were included and randomly divided into training (<i>n</i> = 97) and test (<i>n</i> = 41) cohorts. Radiomics features were extracted from whole lesions and from intralesional subregions generated by clustering analysis. Radiologists’ assessments were performed according to the established MRI grading system. Feature selection was conducted using intraclass correlation coefficient analysis, Pearson correlation filtering, and least absolute shrinkage and selection operator regression. Habitat radiomics, conventional radiomics, and clinicoradiological models were developed using multiple machine learning classifiers. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC), calibration curves, and decision curve analysis. Radiologist performance was assessed with and without model assistance. Model interpretability was explored using SHapley Additive exPlanations (SHAP).</p> Results <p>The habitat radiomics model achieved a test-cohort AUC of 0.838, which was numerically higher than those of the conventional radiomics and clinicoradiological models, although the differences were not statistically significant. Radiologist AUCs ranged from 0.632 to 0.828 without model assistance and from 0.828 to 0.928 with habitat radiomics assistance, with significant improvements mainly observed among junior radiologists (<i>P</i> &lt; 0.05).</p> Conclusion <p>MRI-based habitat radiomics showed feasible performance for differentiating indeterminate Grade 3 nasopharyngeal lesions, although its standalone performance was not statistically superior to comparator models. It may provide complementary assistance for junior radiologists in challenging cases.</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

MR habitat radiomics for differentiating indeterminate (Grade 3) nasopharyngeal lesions: lymphoid hyperplasia versus early-stage nasopharyngeal carcinoma

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月15日

查看全文
  • Zhenhuan Huang,
  • Xuezhao Tu,
  • Qikui You,
  • Dandan Lin,
  • Tingyu Yu,
  • Yueming Li

摘要

Background

To investigate the feasibility of MRI-based habitat radiomics to differentiate indeterminate Grade 3 nasopharyngeal lesions into nasopharyngeal lymphoid hyperplasia or early-stage nasopharyngeal carcinoma, and to explore its complementary value for radiologists in diagnostically ambiguous cases.

Methods

In this retrospective study, 138 patients with Grade 3 nasopharyngeal lesions who underwent contrast-enhanced MRI and endoscopic biopsy were included and randomly divided into training (n = 97) and test (n = 41) cohorts. Radiomics features were extracted from whole lesions and from intralesional subregions generated by clustering analysis. Radiologists’ assessments were performed according to the established MRI grading system. Feature selection was conducted using intraclass correlation coefficient analysis, Pearson correlation filtering, and least absolute shrinkage and selection operator regression. Habitat radiomics, conventional radiomics, and clinicoradiological models were developed using multiple machine learning classifiers. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC), calibration curves, and decision curve analysis. Radiologist performance was assessed with and without model assistance. Model interpretability was explored using SHapley Additive exPlanations (SHAP).

Results

The habitat radiomics model achieved a test-cohort AUC of 0.838, which was numerically higher than those of the conventional radiomics and clinicoradiological models, although the differences were not statistically significant. Radiologist AUCs ranged from 0.632 to 0.828 without model assistance and from 0.828 to 0.928 with habitat radiomics assistance, with significant improvements mainly observed among junior radiologists (P < 0.05).

Conclusion

MRI-based habitat radiomics showed feasible performance for differentiating indeterminate Grade 3 nasopharyngeal lesions, although its standalone performance was not statistically superior to comparator models. It may provide complementary assistance for junior radiologists in challenging cases.