CT人工智能体积参数预测肾癌保留肾单位手术复杂度与T分期
基于两个医学中心600例肾细胞癌患者的增强CT动脉期图像,构建并验证了肾脏与肿瘤自动分割模型(nnU-Net框架),自动提取肿瘤总体积、内生体积及内生比等多维体积参数。结果显示AI体积参数可有效分层预测手术复杂度与病理T分期,并与手工标准具有高度一致性。
要点
- AI自动提取的肿瘤总体积和内生体积可分层预测手术复杂度及T分期。
- AI-内生比分层能力有限,仅可作为术前评分系统的客观补充参数。
- 分割模型对肾脏DSC为0.94、对肿瘤DSC为0.81,肾脏分割精度高于肿瘤。
核心结论
- 基于CT动脉期的AI分割模型可准确自动分割肾脏与肾细胞癌。
- AI多维体积参数能有效预测保留肾单位手术复杂度及病理T分期。
- 该工具为肾细胞癌患者术前个体化评估与手术规划提供可靠定量依据,具有良好的临床转化潜力。
研究设计与样本
- 回顾性收集两个医学中心600例接受保留肾单位手术的肾细胞癌患者CT动脉期图像。
- 采用「全手动精细标注-半自动粗略标注-手动审核」策略构建高质量标注数据集。
- 基于nnU-Net框架训练肾脏和肿瘤分割模型。
- 模型性能以Dice相似系数、归一化表面距离及95百分位豪斯多夫距离进行评估。
- 以组内相关系数分析AI体积参数与手工金标准的一致性,以Kruskal-Wallis检验及Dunn多重比较分析预测能力。
关键影像表现/征象
- 自动提取肿瘤总体积、内生体积及内生比等肿瘤多维体积参数。
- 对内生性肿瘤,内生体积及内生比反映肿瘤向肾实质内生长的程度,影响手术剥离难度。
诊断标准与关键数值
- 肾脏分割DSC为0.94,肿瘤分割DSC为0.81。
- 肾脏NSD为0.93,肿瘤NSD为0.89。
- 肾脏HD95为9.66,肿瘤HD95为11.97。
- AI多维体积参数与手工金标准一致性ICC为0.942至0.986,均P<0.001。
- AI-肿瘤总体积与AI-内生体积可有效分层预测手术复杂度、热缺血时间、手术时长、术中失血量及T分期,均P<0.05。
临床应用与阅片要点
- 术前可利用CT动脉期自动分割并测量肿瘤总体积与内生体积,辅助手术规划和风险分层。
- AI-内生比不宜单独用于预测手术复杂度,可与术前评分系统联合使用。
- 肿瘤分割精度低于肾脏分割,报告相关体积参数时需注意肿瘤边缘分割误差。
局限性
- 本研究为回顾性设计。
- AI-内生比在手术复杂度和T分期分层中缺乏显著能力。
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
BACKGROUND: Based on contrast-enhanced computed tomography (CT) arterial phase images of renal cell carcinoma (RCC), this study aimed to develop an automatic segmentation model for kidneys and RCC, and automatically extract tumor multidimensional volumetric parameters to predict the complexity of nephron-sparing surgery (NSS) and pathological T stage.
MATERIALS AND METHODS: Arterial phase CT images of 600 RCC patients who underwent NSS were retrospectively collected from two medical centers. A high-quality annotated dataset was constructed using the "full manual fine annotation - semi-automatic rough annotation - manual review" strategy, and a kidney and tumor segmentation model was trained based on the nnU-Net framework. Multidimensional volumetric parameters, including total tumor volume (VT), endophytic volume (VE), and endophytic ratio (VE/VT), were extracted. Model performance was evaluated using Dice Similarity Coefficient (DSC), Normalized Surface Distance (NSD), and 95th percentile Hausdorff Distance (HD95). Intraclass Correlation Coefficient (ICC) was used to analyze the consistency between AI-derived multidimensional volumetric parameters and manual gold standard. Kruskal-Wallis test and Dunn's multiple comparisons were applied to assess the predictive ability of these parameters for surgical complexity (warm ischemia time [WIT], surgical duration, intraoperative blood loss) and T stage.
RESULTS: The DSC of the segmentation model for kidneys and tumors was 0.94 and 0.81, respectively; NSD was 0.93 and 0.89, respectively; HD95 was 9.66 and 11.97, respectively. AI-derived multidimensional volumetric parameters showed excellent consistency with the manual gold standard (ICC: 0.942-0.986, all P < 0.001). Stratified analysis revealed that AI-VT and AI-VE could effectively stratify and predict surgical complexity and T stage (all P < 0.05). Although AI-VE/VT lacked significant stratification ability, it could serve as an objective supplementary parameter for preoperative scoring systems.
CONCLUSION: The AI segmentation model constructed in this study can accurately achieve automatic segmentation of kidneys and RCC. The AI multidimensional volumetric parameters can effectively predict NSS complexity and T stage, providing a reliable quantitative tool for preoperative individualized assessment and surgical planning of RCC patients with good clinical translation potential.
引用
- Lin, Su, Huang, et al. CT-based AI multidimensional volumetric parameters for predicting surgical complexity and T stage in renal cell carcinoma undergoing nephron-sparing surgery. Eur J Radiol. 2026. PMID: 42430942. doi:10.1016/j.ejrad.2026.112997
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。