经胸超声心动图造影中卵圆孔未闭的人工智能诊断
Artificial intelligence diagnosis of patent foramen ovale in contrast transthoracic echocardiography.
作者信息
Sheng Yuanyuan, Chen Lixin, Gu Mengjie, Luo Shuyu, Huang Yuxiang, Lin Xiaoxuan, Liu Xiaohua, Liu Qian, Zhong Xiaofang, Peng Guijuan, Li Jian, Shi Bobo, Wang Lin, Xu Jinfeng, Ning Zhaohui, Liu Yingying
机构信息
Shenzhen Medical Ultrasound Engineering Center, Department of Ultrasound, Shenzhen People's Hospital (The Second Clinical Medical College, Jinan University, The First Affiliated Hospital, Southern University of Science and Technology), Shenzhen 518020, China.
School of Information Engineering, Henan University of Science and Technology, Luoyang 471023, China.
出版信息
iScience. 2024 Sep 21;27(11):111012. doi: 10.1016/j.isci.2024.111012. eCollection 2024 Nov 15.
Artificial intelligence (AI) is rarely directly used in patent foramen ovale (PFO) diagnosis. In this study, an AI model was developed to detect the presence of PFO automatically in both contrast transthoracic echocardiography (cTTE) images and videos. The whole intelligent diagnosis neural network framework consists of two functional modules of image segmentation (Unet, = 1866) and image classification (ResNet 101, = 9152). Finally, another test databases, including 20 cTTE videos (4609 cTTE images), was used to compare the RLS classification model accuracy between AI model and different levels of physicians. The Dice similarity coefficient of left chamber segmentation model of cTTE images was 91.41%, the accuracy of PFO-RLS classification model of cTTE images was 83.55%, the accuracy of PFO-RLS classification model of cTTE videos was 90%. Besides, the AI diagnosis time was significantly shorter than doctors (at only 1.3 s).
人工智能（AI）很少直接用于卵圆孔未闭（PFO）的诊断。在本研究中，开发了一种AI模型，用于在对比经胸超声心动图（cTTE）图像和视频中自动检测PFO的存在。整个智能诊断神经网络框架由图像分割（Unet， = 1866）和图像分类（ResNet 101， = 9152）两个功能模块组成。最后，使用另一个测试数据库，包括20个cTTE视频（4609张cTTE图像），来比较AI模型与不同水平医生之间的RLS分类模型准确性。cTTE图像左腔分割模型的Dice相似系数为91.41%，cTTE图像PFO-RLS分类模型的准确率为83.55%，cTTE视频PFO-RLS分类模型的准确率为90%。此外，AI诊断时间明显短于医生（仅为1.3秒）。
Zotero 插件