超声联合影像组学在肾透明细胞癌核分级中的预测价值
林泓钊 , 李宁 , 解婷婷 , 王艺桦 , 张曼 , 张莹 , 陈曦 , 张树华
中国医科大学学报 ›› 2026, Vol. 55 ›› Issue (10) : 915 -920.
超声联合影像组学在肾透明细胞癌核分级中的预测价值
Value of ultrasound combined with radiomics for predicting the nuclear grade of clear cell renal cell carcinoma
目的 探讨超声、影像组学及二者联合在术前预测肾透明细胞癌 (ccRCC) 世界卫生组织/国际泌尿病理学会 (WHO/ISUP) 核分级中的诊断价值。方法 选取华北理工大学附属医院经病理确诊的186例ccRCC患者的临床及超声资料,按7 ∶ 3分为训练集 (131例) 和内部验证集 (55例),同期选取天津医科大学第二医院的62例ccRCC患者作为外部验证集;训练集患者肿物特征用于建立模型,训练集、内部验证集及外部验证集肿物均用于评价模型的诊断效能。应用WHO/ISUP核分级系统将训练集分为低级别组 (Ⅰ、Ⅱ级,85例) 和高级别组 (Ⅲ、Ⅳ级,46例),使用单因素分析筛选有统计学差异的超声特征。使用3D-Slicer软件勾画肿物边界、提取影像组学特征,经组内相关系数检验剔除重复性差的特征后,使用最小绝对收缩和选择算子 (LASSO) 完成影像组学特征降维。对筛选及降维后的超声及影像组学特征行多重共线性诊断,剔除共线性高的特征。通过二元logistic回归建立超声、影像组学及二者联合模型,应用受试者操作特征 (ROC) 曲线分析预测效果。结果 训练集中低级别组与高级别组病灶大小差异有统计学意义 (P < 0.05);经组内相关系数检验、LASSO回归及多重共线性诊断后,最终保留5个影像组学特征。联合预测模型在训练集、内部验证集及外部验证集的AUC分别为0.815、0.732、0.792,展现出最优的诊断效能。结论 3种模型均可有效预测ccRCC核分级,联合模型的诊断效能最高,为术前评估提供了新依据。
Objective To evaluate the diagnostic value of ultrasound,radiomics,and their combination for preoperative prediction of the World Health Organization/International Society of Urological Pathology (WHO/ISUP) nuclear grade in clear cell renal cell carcinoma (ccRCC) . Methods Clinical and ultrasound data from 186 pathologically confirmed ccRCC patients at the Affiliated Hospital of North China University of Science and Technology were retrospectively collected and divided into a training set (n = 131) and an internal validation set (n = 55) at a 7 ∶ 3 ratio. An additional 62 ccRCC patients from the Second Hospital of Tianjin Medical University served as the external validation set. Model development was performed using the training set,and diagnostic performance was assessed in all three cohorts. According to the WHO/ISUP grading system,lesions in the training set were classified as low grade (gradesⅠ-Ⅱ,n = 85) or high grade (grades Ⅲ-Ⅳ,n = 46) . Ultrasound features with significant between-group differences were identified by univariate analysis. Tumor boundaries were segmented using 3D-Slicer,and radiomics features were extracted. Features with poor reproducibility were excluded using the intraclass correlation coefficient (ICC) . The least absolute shrinkage and selection operator (LASSO) regression was then applied for feature selection. Multicollinearity analysis was performed to remove highly correlated ultrasound and radiomics features. Binary logistic regression was used to construct ultrasound,radiomics,and combined models. Predictive performance was evaluated using receiver operating characteristic curve analysis. Results In the training set,lesion size differed significantly between the low-grade and high-grade groups (P < 0.05) . After ICC testing,LASSO feature selection,and multicollinearity analysis,five radiomics features were retained. The combined model demonstrated the best diagnostic performance,with area under the curve values of 0.815,0.732,and 0.792 in the training,internal validation,and external validation sets,respectively. Conclusion All three models effectively predicted ccRCC nuclear grade. The combined ultrasound-radiomics model showed the highest diagnostic performance and may provide additional support for preoperative assessment.
| [1] |
|
| [2] |
Moch H,Cubilla A L,Humphrey P A, |
| [3] |
|
| [4] |
|
| [5] |
|
| [6] |
|
| [7] |
李宁,阚艳敏,王艺桦, |
| [8] |
|
| [9] |
|
| [10] |
|
| [11] |
张岱,赵利辉,王海玲, |
| [12] |
|
河北省医学科学研究课题计划(20271052)
河北省级研究生精品课程、专业学位精品教学案例 (库) 立项建设项目(KCJPZ2023030)
/
| 〈 |
|
〉 |