Endoscopy Center and Endoscopy Research Institute, Zhongshan Hospital, Fudan University, Shanghai, China.
Department of Computer Science, University of California, Irvine, Irvine, California, USA.
Gastrointest Endosc. 2019 Apr;89(4):806-815.e1. doi: 10.1016/j.gie.2018.11.011. Epub 2018 Nov 16.
According to guidelines, endoscopic resection should only be performed for patients whose early gastric cancer invasion depth is within the mucosa or submucosa of the stomach regardless of lymph node involvement. The accurate prediction of invasion depth based on endoscopic images is crucial for screening patients for endoscopic resection. We constructed a convolutional neural network computer-aided detection (CNN-CAD) system based on endoscopic images to determine invasion depth and screen patients for endoscopic resection.
Endoscopic images of gastric cancer tumors were obtained from the Endoscopy Center of Zhongshan Hospital. An artificial intelligence-based CNN-CAD system was developed through transfer learning leveraging a state-of-the-art pretrained CNN architecture, ResNet50. A total of 790 images served as a development dataset and another 203 images as a test dataset. We used the CNN-CAD system to determine the invasion depth of gastric cancer and evaluated the system's classification accuracy by calculating its sensitivity, specificity, and area under the receiver operating characteristic curve.
The area under the receiver operating characteristic curve for the CNN-CAD system was .94 (95% confidence interval [CI], .90-.97). At a threshold value of .5, sensitivity was 76.47%, and specificity 95.56%. Overall accuracy was 89.16%. Positive and negative predictive values were 89.66% and 88.97%, respectively. The CNN-CAD system achieved significantly higher accuracy (by 17.25%; 95% CI, 11.63-22.59) and specificity (by 32.21%; 95% CI, 26.78-37.44) than human endoscopists.
We constructed a CNN-CAD system to determine the invasion depth of gastric cancer with high accuracy and specificity. This system distinguished early gastric cancer from deeper submucosal invasion and minimized overestimation of invasion depth, which could reduce unnecessary gastrectomy.
根据指南，内镜下切除仅应适用于早期胃癌侵犯深度局限于胃黏膜或黏膜下层且无论淋巴结是否受累的患者。基于内镜图像准确预测侵犯深度对于筛选适合内镜下切除的患者至关重要。我们构建了一个基于内镜图像的卷积神经网络计算机辅助检测（CNN-CAD）系统，以确定侵犯深度并筛选适合内镜下切除的患者。
从中山医院内镜中心获取胃癌肿瘤的内镜图像。通过使用最先进的预训练卷积神经网络架构 ResNet50 进行迁移学习，开发了一个基于人工智能的 CNN-CAD 系统。总共 790 张图像作为开发数据集，另外 203 张图像作为测试数据集。我们使用 CNN-CAD 系统来确定胃癌的侵犯深度，并通过计算其灵敏度、特异性和受试者工作特征曲线下面积来评估系统的分类准确性。
CNN-CAD 系统的受试者工作特征曲线下面积为.94（95%置信区间[CI]，.90-.97）。在阈值为.5 时，灵敏度为 76.47%，特异性为 95.56%。总体准确率为 89.16%。阳性和阴性预测值分别为 89.66%和 88.97%。CNN-CAD 系统的准确性（提高 17.25%；95%CI，11.63-22.59）和特异性（提高 32.21%；95%CI，26.78-37.44）均显著高于人类内镜医师。
我们构建了一个 CNN-CAD 系统来确定胃癌的侵犯深度，具有较高的准确性和特异性。该系统能够区分早期胃癌与更深的黏膜下侵犯，并最大限度地减少对侵犯深度的高估，从而减少不必要的胃切除术。
