| 本文已被：浏览 752次 下载 599次 |
码上扫一扫！ |
| 计算机断层扫描影像组学在小胰腺癌筛查中的应用 |
| 张浩1△,付贝1△,孟英豪1,方旭1,边云1,汪军1,邵成伟1,卢明智2* |
|
|
(1. 海军军医大学(第二军医大学)第一附属医院影像医学科, 上海 200433;
2. 海军军医大学(第二军医大学)第一附属医院放疗科, 上海 200433
△共同第一作者
*通信作者) |
|
| 摘要: |
| 目的 开发一种基于腹部CT平扫图像的小胰腺癌（肿瘤最大径≤2 cm）辅助诊断工具。方法 回顾性分析在海军军医大学（第二军医大学）第一附属医院收治并经病理证实的206例小胰腺癌患者和268例胰腺正常者的影像学资料，并按时间顺序分成训练集（2014年1月至2019年12月收治的143例小胰腺癌患者和188例胰腺正常者）和验证集（2020年1月至2021年12月收治的63例小胰腺癌患者和80例胰腺正常者）。由2位影像医学科医师通过nnU-Net自动分割模型在腹部CT平扫图像上对胰腺整体进行自动勾画，提取影像组学特征。依次采用方差分析、Spearman相关分析和ROC曲线进行特征的降维和选择，构建极端梯度提升（XGBoost）预测模型。通过ROC曲线评估XGBoost预测模型的诊断效能，采用决策曲线分析法（DCA）评价模型的临床适用性。结果 206例小胰腺癌的大小为（1.69±0.77）cm。在训练集中，XGBoost预测模型诊断小胰腺癌的AUC值、灵敏度、特异度、阳性预测值和阴性预测值分别为0.99、0.92、0.97、0.91和0.98；在验证集中，AUC值、灵敏度、特异度、阳性预测值和阴性预测值分别为0.99、0.94、0.96、0.93和0.97。DCA分析提示患者可从该模型中受益。结论 基于对腹部CT平扫图像的影像组学分析构建的XGBoost预测模型能准确鉴别小胰腺癌患者和胰腺正常者，有望成为筛查小胰腺癌的辅助工具。 |
| 关键词: 胰腺肿瘤 小胰腺癌 极端梯度提升 影像组学 计算机断层扫描 腹部 自动分割模型 |
| DOI：10.16781/j.CN31-2187/R.20220366 |
| 投稿时间：2022-05-01修订日期：2022-10-09 |
| 基金项目:国家自然科学基金（81871352，82171915，82171930），上海申康医院发展中心临床三年行动计划重大临床研究项目（SHDC2020CR4073），上海市自然科学基金（21ZR1478500），上海市科技创新行动计划医学创新研究项目（21Y11910300），海军军医大学（第二军医大学）第一附属医院“234学科夯基计划”（2020YPT001）． |
|
| Computed tomography radiomics in screening of small pancreatic cancer |
| ZHANG Hao1△,FU Bei1△,MENG Ying-hao1,FANG Xu1,BIAN Yun1,WANG Jun1,SHAO Cheng-wei1,LU Ming-zhi2* |
(1. Department of Radiology, The First Affiliated Hospital of Naval Medical University (Second Military Medical University), Shanghai 200433, China;
2. Department of Radiotherapy, The First Affiliated Hospital of Naval Medical University (Second Military Medical University), Shanghai 200433, China
△Co-first authors.
* Corresponding author) |
| Abstract: |
| Objective To develop an abdominal computed tomography (CT)-based adjunctive diagnostic tool for small pancreatic cancer (maximal tumor diameter ≤ 2 cm). Methods The imaging data of 206 patients with small pancreatic cancer confirmed by pathology and 268 normal controls without known pancreatic diseases who were admitted to The First Affiliated Hospital of Naval Medical University (Second Military Medical University) were retrospectively analyzed. The patients were assigned to training set and validation set in chronological order:143 patients with small pancreatic cancer and 188 normal controls admitted from Jan. 2014 to Dec. 2019 were assigned to the training set; and 63 patients with small pancreatic cancer and 80 normal controls admitted from Jan. 2020 to Dec. 2021 were assigned to the validation set. The whole pancreas was automatically delineated on the abdominal CT images by 2 imaging physicians using the nnU-Net automatic segmentation model to extract radiomics features. Variance analysis, Spearman correlation analysis and receiver operating characteristic (ROC) curve were applied to select features. The diagnostic performance of extreme gradient boosting (XGBoost) prediction model was evaluated by ROC curve, and the clinical applicability of XGBoost prediction model was evaluated by decision curve analysis (DCA). Results The tumor size of 206 patients with small pancreatic cancer was (1.69±0.77) cm. The area under curve (AUC), sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) of the XGBoost prediction model in the training set were 0.99, 0.92, 0.97, 0.91 and 0.98, respectively. The AUC, sensitivity, specificity, PPV and NPV of the XGBoost prediction model in the validation set were 0.99, 0.94, 0.96, 0.93 and 0.97, respectively. DCA analysis showed that patients could benefit from this model. Conclusion The XGBoost prediction model based on radiomics analysis of abdominal CT images can accurately differentiate small pancreatic cancer from normal pancreas. It is expected to be an auxiliary tool for screening small pancreatic cancer. |
| Key words: pancreatic neoplasms small pancreatic cancer extreme gradient boosting radiomics computed tomography abdomen self-configuring method |
