文献速递腹部CT

整合影像病理组学机器学习模型预测胃癌术前化疗病理反应

An Integrated Radiopathomics Machine Learning Model to Predict Pathological Response to Preoperative Chemotherapy in Gastric Cancer.
PubMed 文献 · Acad Radiol · 2024-08-30

回顾性纳入2015年1月至2023年6月青岛大学附属医院151例接受术前化疗及手术切除的胃癌患者,基于治疗前增强CT图像与全切片病理H&E图像提取特征构建影像病理组学融合模型,并联合Borrmann分型建立列线图,在验证队列中列线图预测病理反应的AUC为0.797。

要点

  • 融合影像组学与病理组学特征构建的模型在验证队列预测病理反应AUC为0.789。
  • 联合影像病理组学特征与Borrmann分型的列线图在训练队列AUC为0.880、验证队列AUC为0.797。
  • 决策曲线分析显示列线图可为胃癌患者提供有利的临床获益。

核心结论

  • 影像病理组学列线图能较高精度预测胃癌术前化疗的病理反应,为胃癌个性化治疗提供新工具。

研究设计与样本

  • 回顾性纳入2015年1月至2023年6月青岛大学附属医院151例接受术前化疗及手术切除的胃癌患者,每个患者均有治疗前增强CT图像与病理H&E染色全切片图像,训练队列106例。

诊断标准与关键数值

  • 训练队列106例中提取962个影像组学特征和999个病理组学特征,用13个影像组学特征和5个病理组学特征构建融合影像病理组学模型;融合模型验证队列AUC为0.789;基于影像病理组学特征与Borrmann分型的列线图训练队列AUC为0.880、验证队列AUC为0.797。

临床应用与阅片要点

  • 作者认为影像病理组学列线图能较高精度预测胃癌术前化疗病理反应,为个性化治疗提供新工具。

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: Accurately predicting the pathological response to chemotherapy before treatment is important for selecting the appropriate treatment groups, formulating individualized treatment plans, and improving the survival rates of patients with gastric cancer (GC).

METHODS: We retrospectively enrolled 151 patients diagnosed with GC who underwent preoperative chemotherapy and surgical resection at the Affiliated Hospital of Qingdao University between January 2015 and June 2023. Both pretreatment-enhanced computer technology images and whole slide images of pathological hematoxylin and eosin-stained sections were available for each patient. The image features were extracted and used to construct an ensemble radiopathomics machine learning model. In addition, a nomogram was developed by combining the imaging features and clinical characteristics.

RESULTS: In total, 962 radiomics and 999 pathomics signatures were extracted from 106 patients in the training cohort. A fusion radiopathomics model was constructed using 13 radiomics and 5 pathomics signatures. The fusion model showed favorable performance compared to single-omics models, with an area under the curve (AUC) of 0.789 in the validation cohort. Moreover, a combined radiopathomics nomogram (RPN) was developed based on radiopathomics features and the Borrmann type, which is a classification method for advanced GC according to tumor growth pattern and gross morphology. The RPN showed superior predictive performance in the training (AUC 0.880) and validation cohorts (AUC 0.797). The decision curve analysis showed that RPN could provide favorable clinical benefits to patients with GC.

CONCLUSIONS: RPN was able to predict the pathological response to preoperative chemotherapy with high accuracy, and therefore provides a novel tool for personalized treatment of GC.

来源:Acad Radiol,查看原文。版权归原作者所有。

引用

  1. Song Y, Liu S, Liu X, et al. An Integrated Radiopathomics Machine Learning Model to Predict Pathological Response to Preoperative Chemotherapy in Gastric Cancer. Acad Radiol. 2024. PMID: 39214816. doi:10.1016/j.acra.2024.08.014

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。