文献速递泌尿生殖腹部CT

基于多能谱双能CT图像、整合栖息地影像组学与视觉Transformer深度学习用于膀胱癌肌层浸润术前预测的可解释模型

Multi-DECT Image-based Interpretable Model Incorporating Habitat Radiomics and Vision Transformer Deep Learning for Preoperative Prediction of Muscle Invasion in Bladder Cancer.
PubMed 文献 · Acad Radiol · 2025-08-30

回顾性分析200例膀胱癌患者(训练集140例、测试集60例),基于40、70、100 keV虚拟单能图像构建传统影像组学、栖息地、ResNet 18、ViT及融合模型,其中栖息地-ViT模型表现最优,训练集AUC 0.997(95%CI 0.992, 1.000)、测试集AUC 0.892(95%CI 0.814, 0.971),加入DECT定量参数未提升性能,SHAP方法可解释模型决策过程。

要点

  • 栖息地-ViT融合模型在测试集预测膀胱癌肌层浸润的AUC为0.892(95%CI 0.814, 0.971)。
  • 在40、70、100 keV虚拟单能图像上构建的栖息地-ViT模型优于其他单一模型。
  • 加入DECT定量参数并未改善模型预测性能。

核心结论

  • 整合栖息地影像组学与ViT深度学习的多能谱双能CT可解释模型有望用于术前预测膀胱癌肌层浸润状态,为个体化治疗方案制定和预后评估提供有价值的信息。

研究设计与样本

  • 回顾性研究,纳入200例膀胱癌患者,按7:3分为训练集140例和测试集60例;基于双能CT的40、70、100 keV虚拟单能图像构建传统影像组学、栖息地、ResNet 18、ViT及融合模型;采用AUC、校准曲线、决策曲线分析、净重分类指数和综合判别改善指数评价模型,并以SHAP方法解释最优模型。

诊断标准与关键数值

  • 栖息地-ViT模型训练集AUC 0.997(95%CI 0.992, 1.000)、测试集AUC 0.892(95%CI 0.814, 0.971);决策曲线分析和校准曲线显示该模型提供良好的净获益并表现出强校准。

临床应用与阅片要点

  • 作者认为,基于多能谱双能CT、整合栖息地影像组学与ViT深度学习构建的可解释模型有望用于术前预测膀胱癌肌层浸润状态;SHAP方法可阐明模型预测结局的决策过程。

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

RATIONALE AND OBJECTIVES: The research aims to evaluate the effectiveness of a multi-dual-energy CT (DECT) image-based interpretable model that integrates habitat radiomics with a 3D Vision Transformer (ViT) deep learning (DL) for preoperatively predicting muscle invasion in bladder cancer (BCa).

MATERIALS AND METHODS: This retrospective study analyzed 200 BCa patients, who were divided into a training cohort (n=140) and a test cohort (n=60) in a 7:3 ratio. Univariate and multivariate analyses were performed on the DECT quantitative parameters to identify independent predictors, which were subsequently used to develop a DECT model. The K-means algorithm was employed to generate habitat sub-regions of BCa. Traditional radiomics (Rad) model, habitat model, ResNet 18 model, ViT model, and fusion models were constructed from the 40, 70, and 100 keV virtual monochromatic images (VMIs) in DECT. The evaluation of all models used the area under the receiver operating characteristic curve (AUC), calibration curve, decision curve analysis (DCA), net reclassification index (NRI), and integrated discrimination improvement (IDI). The SHAP method was employed to interpret the optimal model and visualize its decision-making process.

RESULTS: The Habitat-ViT model demonstrated superior performance compared to other single models, achieving an AUC of 0.997 (95% CI 0.992, 1.000) in the training cohort and 0.892 (95% CI 0.814, 0.971) in the test cohort. The incorporation of DECT quantitative parameters did not improve the performance. DCA and calibration curve assessments indicated that the Habitat-ViT model provided a favorable net benefit and demonstrated strong calibration. Furthermore, SHAP clarified the decision-making processes underlying the model's predicted outcomes.

CONCLUSION: Multi-DECT image-based interpretable model that integrates habitat radiomics with a ViT DL holds promise for predicting muscle invasion status in BCa, providing valuable insights for personalized treatment planning and prognostic assessment.

来源:Acad Radiol,查看原文。版权归原作者所有。

引用

  1. Du C, Wei W, Hu M, et al. Multi-DECT Image-based Interpretable Model Incorporating Habitat Radiomics and Vision Transformer Deep Learning for Preoperative Prediction of Muscle Invasion in Bladder Cancer. Acad Radiol. 2025. PMID: 40887351. doi:10.1016/j.acra.2025.08.018

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。