Vu Trung Kien, Hollenbach Marcus, Veldhuizen Gregory Patrick, Saldanha Oliver Lester, Garbe Jakob, Rosendahl Jonas, Krug Sebastian, Michl Patrick, Feisthammel Jürgen, Karlas Thomas, Hampe Jochen, Hoffmeister Albrecht, Kather Jakob Nikolas
Division of Gastroenterology, Medical Department II, University of Leipzig Medical Center, Leipzig, Germany,
Division of Gastroenterology, Medical Department II, University of Leipzig Medical Center, Leipzig, Germany.
Digestion. 2025;106(4):287-302. doi: 10.1159/000543049. Epub 2024 Dec 13.
The accurate distinction between benign and malignant biliary strictures (BSs) poses a significant challenge. Despite the use of bile duct biopsies and brush cytology via endoscopic retrograde cholangiopancreaticography (ERCP), the results remain suboptimal. Single-operator cholangioscopy can enhance the diagnostic yield in BS, but its limited availability and high costs are substantial barriers. Convolutional neural network-based systems may improve the diagnostic process and enhance reproducibility. Therefore, we assessed the feasibility of using deep learning to differentiate BS using fluoroscopy images during ERCP.
We conducted a retrospective review of adult patients (n = 251) from three university centers in Germany (Leipzig, Dresden, Halle) who underwent ERCP. We developed and evaluated a deep learning-based model using fluoroscopy images. The performance of the classifier was evaluated by measuring the area under the receiver operating characteristic curve (AUROC), and we utilized saliency map analyses to understand the decision-making process of the model.
In cross-validation experiments, malignant BSs were detected with a mean AUROC of 0.89 ± 0.03. The test set of the Leipzig cohort demonstrated an AUROC of 0.90. In two independent external validation cohorts (Dresden, Halle), the deep learning-based classifier achieved an AUROC of 0.72 and 0.76, respectively. The artificial intelligence model's predictions identified plausible characteristics within the fluoroscopy images.
By using a deep learning model, we were able to discriminate malignant BS from benign biliary conditions. The application of artificial intelligence enhances the diagnostic yield of malignant BS and should be validated in a prospective design.
准确区分良性和恶性胆管狭窄（BS）是一项重大挑战。尽管通过内镜逆行胰胆管造影（ERCP）进行胆管活检和刷检细胞学检查，但结果仍不尽人意。单操作者胆管镜检查可提高BS的诊断率，但其可用性有限且成本高昂，构成了重大障碍。基于卷积神经网络的系统可能会改善诊断过程并提高可重复性。因此，我们评估了在ERCP期间使用深度学习通过荧光透视图像区分BS的可行性。
我们对来自德国三个大学中心（莱比锡、德累斯顿、哈雷）接受ERCP的成年患者（n = 251）进行了回顾性研究。我们使用荧光透视图像开发并评估了一个基于深度学习的模型。通过测量受试者工作特征曲线（AUROC）下的面积来评估分类器的性能，并且我们利用显著性图分析来了解模型的决策过程。
在交叉验证实验中，检测恶性BS的平均AUROC为0.89±0.03。莱比锡队列的测试集显示AUROC为0.90。在两个独立的外部验证队列（德累斯顿、哈雷）中，基于深度学习的分类器的AUROC分别为0.72和0.76。人工智能模型的预测识别出了荧光透视图像中的合理特征。
通过使用深度学习模型，我们能够将恶性BS与良性胆管疾病区分开来。人工智能的应用提高了恶性BS的诊断率，应在前瞻性设计中进行验证。
