A Multi-stage Segmentation Method for Tongue Ecchymosis.

Tongue diagnosis is an important component of traditional Chinese medicine (TCM), in which tongue ecchymosis is the main diagnostic basis for the blood stasis syndrome of TCM. Most of the existing methods are unsupervised and cannot accurately segment tongue ecchymosis. In this paper, we propose a multi-stage segmentation method for tongue ecchymosis. We first employ an object detection model for rough localization of tongue ecchymosis, and then use the unsupervised clustering and the watershed transform for rough segmentation and fine segmentation of tongue ecchymosis respectively. To the best of our knowledge, we are the first to combine machine learning and deep learning to segment tongue ecchymosis. Experimental results show that the tongue ecchymoses obtained by our method are more similar to the real tongue ecchymoses compared with the existing methods, and the Intersection-over-Union (IoU) is improved by 0.12 compared with the latest method.Clinical Relevance-Tongue ecchymosis obtained by this paper is the main diagnostic basis for the blood stasis syndrome of TCM.