Study and Application on the Data Curation of Medical Images Based on
Image Classification Technology
HAN Xiao1, GU Zongyun1,2, ZHAO Shibo1, SONG Liangliang3, WANG Qian3, LI Chuanfu2,3
1. College of Medical Information Engineering, Anhui University of Chinese Medicine, Hefei Anhui 230012, China;
2. Institue of Artificial Intelligence of Hefei Comprehensive National Science Center, Anhui Artificial Intelligence Laboratory,
Hefei Anhui 230088, China; 3. Department of Imaging Center, First Affiliated Hospital of Anhui University of Traditional Chinese
Medicine, Hefei Anhui 230031, China
Abstract：Objective To explore the feasibility of using deep learning image classification method to accurately classify each X-ray
images in the process of data curation. Methods The desensitized X-ray data uploaded to the Anhui Imaging Cloud Platform by
different types of equipment from more than 80% of medical institutions in Anhui Province were included in this study. A total
of 25 common X-ray examination items were selected. Two imaging technicians annotated 12857 image data obtained from the
data curation platform by referring to the names of examination items in the radiology information system. The labeled data were
divided into training set, validation set and testing set by the ratio of 7∶1∶2. The training set contained 9006 images, validation set
contained 1279 images,and testing set contained 2572 images. A deep learning network was built based on ResNet50, the training
set and validation set were used for model training. The receiver operating characteristic curve (ROC) curve and sensitivity of X-ray
items were used as evaluation indicators. Results The test results of the testing set showed that the majority of images were correctly
classified, the area under the average ROC curve of image classification were 99.94%, and the sensitivity were 98.05%±5.68%, but
there were still a small number of image were inaccurate. Conclusion In the process of medical imaging data curation, the method
based on image classification can correctly classify the most of X-ray images. However, a small number of non-standard images
cannot be correctly classified due to the non-standard shooting images. So it needs to be further classified in combination with the
examination items name information of the radiology information system.
韩啸1，谷宗运1,2，赵士博1，宋亮亮3，王倩3，李传富2,3. 基于图像分类技术在医学影像数据
治理过程中的研究与应用[J]. 中国医疗设备, 2023, 38(4): 78-83.
HAN Xiao1, GU Zongyun1,2, ZHAO Shibo1, SONG Liangliang3, WANG Qian3, LI Chuanfu2,3. Study and Application on the Data Curation of Medical Images Based on
Image Classification Technology. China Medical Devices, 2023, 38(4): 78-83.
