Tongue image datasets and performance evaluation

Abstract

With the development of computer medicine, data has become an extremely important resource, and the construction and research of dataset is an important means to promote the development and innovation of medicine. Because there are few tongue image datasets and most tongue body image datasets are in a non-public state, the research of medical related algorithms is hindered to some extent, especially in computer vision and medical artificial intelligence. The constructed tongue image data set provides a large number of tongue images and related clinical information, which can be used to train computer medical models such as tongue image detection and tongue segmentation. A Tongue dataset based on deep learning was established by selecting and labeling the collected tongue images manually. The datasets contain 2,021 tongue images, which are randomly divided into training sets, verification sets and test sets according to the proportion, and the dataset performance evaluation experiments are carried out on the existing algorithms. Testing on convolutional neural network models such as YOLO detection and tongue segmentation, the average accuracy of detection and segmentation Miou value can reach 97.0% and 99.95%, respectively. The experimental results show that the data set can obtain high accuracy in detection and segmentation algorithms, and can be further applied to related research in the medical field. Tongue dataset data set can be downloaded at the following url: https://github.com/xiaohuomiao12/datasets.git.