Department of Gastroenterology, The Third Xiangya Hospital of Central South University, Changsha, Hunan, China.
Hunan Key Laboratory of Nonresolving Inflammation and Cancer, Central South University, Changsha, Hunan, China.
Med Phys. 2023 Jul;50(7):4197-4205. doi: 10.1002/mp.16390. Epub 2023 Apr 10.
Early detection of solid pancreatic masses through contrast-enhanced harmonic endoscopic ultrasound (CH-EUS) is important. But CH-EUS is difficult to learn.
To design a deep learning-based CH-EUS diagnosis system (CH-EUS MASTER) for real-time capture and segmentation of solid pancreatic masses and to verify its value in the training of pancreatic mass identification under endoscopic ultrasound (EUS).
We designed a real-time capture and segmentation model for solid pancreatic masses and then collected 4530 EUS images of pancreatic masses retrospectively, used for training and testing of this model at a ratio of 8:2. The model is loaded into the EUS host computer to establish the CH-EUS MASTER system. A crossover trial was then conducted to evaluate the model's value in EUS trainee training by successfully conducting two groups of EUS trainees in model learning and trainer-guided training. The intersection over union (IoU) and the time to find the lesion were used to evaluate the model performance metrics, and the Mann-Whitney test was used to compare the IoU and the time to find the lesion in different groups of subjects. Paired t-test was used to compare the effects before and after training. When α ≤ 0.05, it is considered to have a significant statistical difference.
The model test showed that the model successfully captured and segmented the pancreatic solid mass region in 906 EUS images. The real-time capture and segmentation model had a Dice coefficient of 0.763, a recall rate of 0.941, a precision rate of 0.642, and an accuracy of 0.842 (when the threshold is set to 0.5), and the median IoU of all cases was 0.731. For the AI training effect, the average IoU of eight trainees improved from 0.80 to 0.87 (95% CI, 0.032-0.096; p = 0.002). The average time for identifying lesions in the pancreatic body and tail improved from 22.75 to 17.98 s (95% CI, 3.664-5.886; p < 0.01). The average time for identifying lesions in the pancreatic head and uncinate process improved from 34.21 to 25.92 s (95% CI, 7.661-8.913; p < 0.01).
CH-EUS MASTER can provide an effect equivalent to trainer guidance in training pancreatic solid mass identification and segmentation under EUS.
通过对比增强谐波内镜超声（CH-EUS）早期检测实体胰腺肿块很重要。但 CH-EUS 很难学习。
设计一种基于深度学习的 CH-EUS 诊断系统（CH-EUS MASTER），用于实时捕获和分割实体胰腺肿块，并验证其在 EUS 下胰腺肿块识别培训中的价值。
我们设计了一个用于实体胰腺肿块的实时捕获和分割模型，然后回顾性地收集了 4530 个胰腺肿块的 EUS 图像，用于模型训练和测试的比例为 8:2。将模型加载到 EUS 主机中，建立 CH-EUS MASTER 系统。然后进行交叉试验，通过两组 EUS 学员在模型学习和导师指导培训中的成功学习，评估模型在 EUS 学员培训中的价值。使用交并比（IoU）和找到病变的时间来评估模型性能指标，使用 Mann-Whitney 检验比较不同组受试者的 IoU 和找到病变的时间。使用配对 t 检验比较培训前后的效果。当 α≤0.05 时，认为具有显著的统计学差异。
模型测试表明，该模型成功地捕获并分割了 906 个 EUS 图像中的胰腺实体肿块区域。实时捕获和分割模型的 Dice 系数为 0.763，召回率为 0.941，准确率为 0.642，准确率为 0.842（当阈值设置为 0.5 时），所有病例的中位数 IoU 为 0.731。对于 AI 培训效果，八名学员的平均 IoU 从 0.80 提高到 0.87（95%CI，0.032-0.096；p=0.002）。识别胰腺体尾部病变的平均时间从 22.75 秒提高到 17.98 秒（95%CI，3.664-5.886；p<0.01）。识别胰头部和钩突病变的平均时间从 34.21 秒提高到 25.92 秒（95%CI，7.661-8.913；p<0.01）。
CH-EUS MASTER 可以在 EUS 下提供与导师指导相当的胰腺实体肿块识别和分割培训效果。
