Deep Learning-Based Tongue Image Segmentation Research

With the development of modern medical imaging and artificial intelligence, tongue image segmentation has become a critical step in intelligent tongue diagnosis systems. The quality of segmentation directly impacts subsequent diagnoses, necessitating precise tongue segmentation to reduce interference from irrelevant regions and enhance diagnostic accuracy. Currently, both traditional tongue image segmentation methods and certain deep learning approaches have yet to attain the required levels of accuracy and speed for effective tongue image segmentation. In this paper, we conducted research based on deep learning, using the UNet model for medical image segmentation as a foundation. We combined Convolutional Neural Networks (CNN) with Transformers equipped with attention mechanisms. Additionally, we introduced the novel neural network operator called Involution. As a result of these efforts, we achieved a remarkable 99.05 percent accuracy in segmenting tongue images, primarily collected by Prof. Zhao's team. We applied this segmentation method to the tongue diagnostic process for conditions such as Qi deficiency and tongue pain. When applied to the tongue diagnosis process in the tongue image classification of Qi deficiency and calm quality tongue images, the classification effect is improved by about 10 percent compared with the classification effect without tongue image segmentation. When applied to the tongue image classification for Qi deficiency and calm quality tongue tongue images in the tongue diagnostic process,our method improved the classification accuracy by approximately 10 percent compared to classification without tongue image segmentation.