唐迁, 杜博, 恽爽, 高莉, 吴爽, 张超, 兰猛, 陈紫业, 李亮, 查云飞, 张良培, 李平湘. COVID-19 CT影像智能诊断系统[J]. 武汉大学学报 ( 信息科学版), 2020, 45(6): 846-853. DOI: 10.13203/j.whugis20200225
引用本文: 唐迁, 杜博, 恽爽, 高莉, 吴爽, 张超, 兰猛, 陈紫业, 李亮, 查云飞, 张良培, 李平湘. COVID-19 CT影像智能诊断系统[J]. 武汉大学学报 ( 信息科学版), 2020, 45(6): 846-853. DOI: 10.13203/j.whugis20200225
TANG Qian, DU Bo, YUN Shuang, GAO Li, WU Shuang, ZHANG Chao, LAN Meng, CHEN Ziye, LI Liang, ZHA Yunfei, ZHANG Liangpei, LI Pingxiang. Intelligent Diagnosis System Based on COVID-19 CT Images[J]. Geomatics and Information Science of Wuhan University, 2020, 45(6): 846-853. DOI: 10.13203/j.whugis20200225
Citation: TANG Qian, DU Bo, YUN Shuang, GAO Li, WU Shuang, ZHANG Chao, LAN Meng, CHEN Ziye, LI Liang, ZHA Yunfei, ZHANG Liangpei, LI Pingxiang. Intelligent Diagnosis System Based on COVID-19 CT Images[J]. Geomatics and Information Science of Wuhan University, 2020, 45(6): 846-853. DOI: 10.13203/j.whugis20200225

COVID-19 CT影像智能诊断系统

Intelligent Diagnosis System Based on COVID-19 CT Images

  • 摘要: 基于肺部CT(computed tomography)影像的人工智能诊断是针对新型冠状病毒肺炎(coronavirus disease 2019,COVID-19)的有效辅助诊断方法之一。使用神经网络以及数字图像处理等技术,设计了一个基于切片内和切片间注意力机制的轻量级COVID-19分类模型,在此基础上开发了集早期筛查、病变评估、病灶分割功能和肺部及病灶像素分布直方图等功能于一体的COVID-19智能诊断系统。通过从武汉大学人民医院采集了247名COVID-19病患、152名其他肺炎患者和92名健康者的肺部CT图像,并制作为训练数据集用于网络训练。实验结果显示,提出的方法在验证集上的筛查任务和病变评估任务上的准确率分别达到88.63%和89.65%,算法模型中每人平均诊断时间缩短到0.4 s,系统具有重要的应用价值。

     

    Abstract: Chest CT(computed tomography) imaging diagnosis is one of the main diagnostic methods for the coronavirus disease 2019(COVID-19). Deep learning technologies such as convolutional neural networks are widely used in medical image processing because of their powerful nonlinear modeling capabilities.A neural network and digital image processing technology is used to design a lightweight COVID-19 classification model based on intra-volume and inter-volume attention mechanisms. Based on this model, we developed a new COVID-19 intelligent diagnosis system with a set of diagnostic functions, lesion segmentation functions and lung and pixel distribution histogram functions. We collected CT images of lungs from 247 patients with COVID-19, 152 other patients with pneumonia and 92 healthy people from the People's Hospital of Wuhan University and made them as training data sets for network training. The experimental results show that the accuracy of our proposed method on the screening task and the degree grading task on the validation set reached 88.63% and 89.65%, respectively, and the average diagnosis time per person was shortened to 0.4 seconds in the algorithm module, which has important application value.

     

/

返回文章
返回