双能CT不同低能级虚拟单色图像对预测膀胱癌肌层浸润的影像组学模型的影响
研究纳入127例病理证实的膀胱癌患者(肌层浸润49例、非肌层浸润78例),按7:3随机分为训练集和测试集,在静脉期40、50、60、70 keV虚拟单色图像上提取影像组学特征,比较单能分析与多能分析联合支持向量机模型预测肌层浸润状态的效能。测试集中多能模型诊断效能最佳,AUC为0.917,敏感度0.800、特异度0.833、准确率0.821、精确率0.750;单能模型预测肌层浸润状态的AUC和敏感度较低。
要点
- 多能级虚拟单色图像联合的影像组学模型在测试集预测膀胱癌肌层浸润的AUC达0.917。
- 多能模型测试集敏感度0.800、特异度0.833、准确率0.821、精确率0.750。
- 单能模型预测肌层浸润状态的AUC和敏感度较多能模型低。
核心结论
- 联合不同能量虚拟单色图像信息的多能模型在术前预测膀胱癌肌层浸润状态方面表现出更优性能。
研究设计与样本
- 回顾性研究纳入127例病理证实的膀胱癌患者(肌层浸润49例、非肌层浸润78例),按7:3随机分为训练集和测试集,在静脉期40、50、60、70 keV虚拟单色图像上提取特征,采用递归特征消除和LASSO筛选特征,以支持向量机建立模型。
诊断标准与关键数值
- 测试集中多能模型AUC 0.917、敏感度0.800、特异度0.833、准确率0.821、精确率0.750;单能模型预测肌层浸润状态的AUC和敏感度较多能模型低。
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: The purpose of this study was to investigate the impact of different low-energy virtual monochromatic images (VMIs) in dual-energy CT on the performance of radiomics models for predicting muscle invasive status in bladder cancer (BCa).
MATERIALS AND METHODS: A total of 127 patients with pathologically proven muscle-invasive BCa (n = 49) and non-muscle-invasive BCa (n = 78) were randomly allocated into the training and test cohorts at a ratio of 7:3. Feature extraction was performed on the venous phase images reconstructed at 40, 50, 60 and 70-keV (single-energy analysis) or in combination (multi-energy analysis). Recursive feature elimination (RFE) and the least absolute shrinkage and selection operator (LASSO) were employed to select the most relevant features associated with BCa. Models were built using a support vector machine (SVM) classifier. Diagnostic performance was assessed through receiver operating characteristic curves, evaluating sensitivity, specificity, accuracy, precision, and the area-under-the curve (AUC) values.
RESULTS: In the test cohort, the multi-energy model achieved the best diagnostic performance with AUC, sensitivity, specificity, accuracy, and precision of 0.917, 0.800, 0.833, 0.821, and 0.750, respectively. Conversely, the single-energy model exhibited lower AUC and sensitivity in predicting the muscle invasion status.
CONCLUSIONS: By combining information from VMIs of various energies, the multi-energy model displays superior performance in preoperatively predicting the muscle invasion status of bladder cancer.
引用
- Hu M, Wei W, Zhang J, et al. Impact of virtual monochromatic images of different low-energy levels in dual-energy CT on radiomics models for predicting muscle invasion in bladder cancer. Abdom Radiol (NY). 2024. PMID: 38937340. doi:10.1007/s00261-024-04459-6
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。