Oral U-Net: a network model for oral ulcer lesion segmentation

Oral ulcers are a common lesion and are also a complication of various diseases, including oral cancer. In clinical practice, the diagnosis and analysis of oral ulcers are crucial. Accurately identifying and segmenting the ulcer regions in oral images aids in the diagnosis and treatment of multiple diseases. In this paper, we propose a network model for oral ulcer lesion segmentation, called Oral U-Net. We improved upon previous methods (HF-UNet) by removing some modules and replacing the bottom layer of the HF-UNet model with the HIFA module from the I2U-Net model. The Oral U-Net model performed excellently on the oral ulcer dataset. Experimental results show that the mDice is 81.19%andACC is 97.25%, both surpassing other mainstream models. We further validated the model using the colonoscopy polyp segmentation dataset, and the results showed that the mDice is 93.41% and ACC is 99.17%, again achieving the current best performance.