| 林冰冰,尹平,郑飞,洪楠.基于平扫CT影像组学预测急性胰腺炎复发[J].中国医学影像技术,2025,41(5):749~752 |
| 基于平扫CT影像组学预测急性胰腺炎复发 |
| Non-contrast CT radiomics for predicting recurrence of acute pancreatitis |
| 投稿时间：2024-11-22 修订日期：2025-04-28 |
| DOI：10.13929/j.issn.1003-3289.2025.05.012 |
| 中文关键词: 胰腺炎 复发 体层摄影术,X线计算机 影像组学 |
| 英文关键词:pancreatitis recurrence tomography, X-ray computed radiomics |
| 基金项目:国家自然科学基金(81971575)、北京联影智能影像技术研究院基金项目(CRIBJQY202105)。 |
|
| 摘要点击次数: 62 |
| 全文下载次数: 16 |
| 中文摘要: |
| 目的 观察基于平扫CT影像组学预测急性胰腺炎(AP)复发的价值。方法 回顾性分析356例首发AP患者,根据症状完全/近乎消失3个月后有无复发将其分为复发组(n=78)与未复发组(n=278),同时按6∶4比例划分训练集(n=213)与测试集(n=143)。参考门静脉期图像对其中116例接受增强CT检查者于平扫CT中手动勾画胰腺实质ROI;基于其余240例平扫CT数据设计SegResNet模型以自动分割胰腺实质ROI,提取并筛选其最优影像组学特征,并以YeoJohnson变换及Bagging决策树建立影像组学模型。绘制受试者工作特征曲线,计算曲线下面积(AUC),评估所获模型预测AP复发的效能。结果 共于胰腺实质ROI中提取2 264个影像组学特征,最终筛选得到4个最优特征,以之构建的影像组学模型预测训练集及测试集AP复发的AUC分别为0.887及0.889。结论 基于平扫CT影像组学可有效预测AP复发。 |
| 英文摘要: |
| Objective To observe the value of non-contrast CT radiomics for predicting recurrence of acute pancreatitis (AP). Methods Totally 356 patients with first-episode AP were retrospectively enrolled. The patients were categorized into recurrence group (n=78) and non-recurrence group (n=278) based on whether recurrence after 3 months of complete/near disappearance of symptoms, also divided into training set (n=213) and test set (n=143) at the ratio of 6 ∶ 4. For 116 cases who underwent contrast-enhanced CT, taken portal venous phase images as references, ROI of pancreatic parenchyma was manually delineated on non-contrast CT, while SegResNet segmentation model was used for automatic segmentation on non-contrast CT images for the rest 240 cases. The optimal radiomics features were extracted and selected to construct a radiomics model based on YeoJohnson transformer and Bagging decision tree. The receiver operating characteristic curve was drawn, and the area under the curve (AUC) was calculated to evaluate the efficacy of the obtained model for predicting AP recurrence. Results Totally 2 264 radiomics features were extracted from ROI of pancreatic parenchyma, and finally 4 optimal features were screened. The AUC of radiomics model for predicting recurrence was 0.887 and 0.889 in training set and test set, respectively. Conclusion Non-contrast CT radiomics could be used to effectively predict recurrence of AP. |
| 查看全文 查看/发表评论 下载PDF阅读器 |
|
|
|
