TU-Net - A Precise Network for Tongue Segmentation.

Tongue diagnosis is a valuable clinical experience accumulated by Traditional Chinese Medicine (TCM) through long-term research. In TCM, tongue body reflects the most sensitive indicators of the physiological function and pathological changes, which can help doctors to determine the strategy of syndrome differentiation. Therefore, tongue segmentation is particularly important for intelligent tongue diagnosis. With the development of Convolutional Neural Network (CNN) in image processing, some popular networks such as U-shape Network (U-Net), Fully Convolutional Networks (FCN) and their variants have been used in medical image segmentation. It is challenging to segment tongue because the pixels of human lips, chin, and other parts in the tongue images are the same as the tongue. In this paper, we propose an end-to-end network called Tongue U-Net (TU-Net) which combines the classical U-Net structure with Squeeze-and-Excitation (SE) block, Dense Atrous Convolution (DAC) block and Residual Multi-kernel Pooling (RMP) block. The model is inspired from Squeeze-and-Excitation Networks (SENet) and Context Encoder Network (CE-Net) that can capture more useful information. Applied to a tongue dataset with 300 images, TU-Net performs better than the four segmentation methods (FCN, U-Net, Attention U-Net and U2 -Net) in the evaluation of Dice coefficient, Intersection over Union and Hausdorff distance.