基于2019 Bosniak分类的CT预测模型开发与验证:用于肾囊性肿瘤精准分层的一项多中心回顾性研究
该多中心回顾性研究基于2019 Bosniak分类开发并验证了简化CT预测模型(SCTM)用于鉴别肾囊性肿块良恶性,并建立XGBoost模型区分惰性与侵袭性肾囊性肿块。SCTM在验证集及三个测试集AUC分别为0.99、0.89、0.96、0.95,XGBoost模型在验证集、测试1集及验证与测试合并集AUC分别为0.92、0.80、0.87。
要点
- SCTM仅纳入分隔数量与凸出角度两项2019 Bosniak分类特征,验证集AUC 0.99。
- SCTM在测试1、测试2、测试3集AUC分别为0.89、0.96、0.95。
- XGBoost模型整合年龄、性别、BMI、位置、大小及SCTM概率,验证集AUC 0.92。
核心结论
- SCTM是一种简化、准确、无创的CT模型,可有效鉴别肾囊性肿块良恶性;XGBoost模型进一步细化惰性肾囊性肿块的识别,优化治疗并减少不必要手术。
研究设计与样本
- 多中心回顾性研究,纳入2013年1月至2022年12月四家机构经病理确诊的1138例肾囊性肿块患者,训练集655例、内部验证集165例、外部测试集146例、69例、103例,由两名放射科医师按2019 Bosniak分类评估,采用多变量逻辑回归与XGBoost建模。
关键影像表现/征象
- SCTM由分隔数量与凸出角度组合构成。
- XGBoost模型整合年龄、性别、BMI、位置、大小及SCTM概率。
诊断标准与关键数值
- SCTM在验证集AUC 0.99、测试1集AUC 0.89、测试2集AUC 0.96、测试3集AUC 0.95。
- XGBoost模型在验证集AUC 0.92、测试1集AUC 0.80、验证与测试合并集AUC 0.87。
- 1138例中良性肾囊性肿块710例(62.39%),恶性428例(37.61%)。
临床应用与阅片要点
- SCTM利用2019 Bosniak分类关键特征鉴别肾囊性肿块良恶性。
- 联合XGBoost模型可细化惰性与侵袭性肾囊性肿块的识别,辅助制定个体化治疗决策、介入干预并减少不必要手术。
局限性
- 原文未报告局限性。
以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。
英文原文(摘要)· 点击展开
OBJECTIVE: To develop and validate a simplified CT prediction model (SCTM) based on the 2019 Bosniak classification for predicting malignancy in cystic renal masses (CRMs), and to create a secondary model to distinguish indolent from aggressive CRMs.
MATERIALS AND METHODS: This retrospective study included the consecutive patients who were pathologically confirmed to have malignant and benign CRMs at four institutions from January 2013 to December 2022. Two radiologists evaluated all CRMs using the Bosniak classification. Multivariable logistic regression and XGBoost models were used to create prediction models. The models were developed, and validated in an independent internal cohort and four independent external cohorts. Model performance was assessed for differentiation ability and clinical utility.
RESULTS: In total, 1138 CRM patients were analyzed across training (n = 655), internal validation (n = 165), and external tests (n = 146, n = 69, n = 103). Benign and malignant CRMs accounted for 710 (62.39%) and 428 (37.61%), respectively. The SCTM, which combined the number of septa and the angle of convex protrusion, showed excellent differentiation in the validation (AUC 0.99), test 1 (AUC 0.89), test 2 (AUC 0.96), and test 3 (AUC 0.95) sets. The XGBoost model, which integrated age, sex, BMI, location, size, and SCTM probability to differentiate indolent from aggressive CRMs, showed good performance with AUCs of 0.92, 0.80, and 0.87 in the validation, test 1, and combined validation and test sets.
CONCLUSION: SCTM was a simplified, accurate, non-invasive CT model that effectively differentiated benign from malignant CRMs, and the XGBoost model refined the identification of indolent CRMs, optimizing treatments and reducing unnecessary surgeries.
KEY POINTS: Question Can an SCTM accurately predict malignancy in CRMs and identify aggressive lesions to guide personalized clinical decisions? Findings The SCTM and XGBoost models achieved high diagnostic accuracy in distinguishing malignant and aggressive CRMs across internal and external cohorts. Clinical relevance The SCTM, utilizing key 2019 Bosniak features, differentiates benign from malignant CRMs. Combined with the XGBoost model, refining indolent vs aggressive CRM identification, this achieves precise renal cystic tumor stratification, aiding in tailored treatment decisions, interventions, and reducing unnecessary surgeries.
引用
- Kang Q, Li N, Chen C, et al. Development and validation of a CT prediction model for precise stratification of renal cystic tumors: a multicenter retrospective study based on the BOSNIAK classification. Eur Radiol. 2025. PMID: 40540012. doi:10.1007/s00330-025-11720-z
供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。