Method of rectal tumor segmentation based on ResUnet++

Abstract

Rectal cancer is one of the most common malignant tumors. Electronic cross section examination (CT) is used as a screening tool in the diagnosis of rectal cancer. The application of computer aided diagnosis technology to help doctors distinguish between benign and malignant tumors in rectal CT images is of great significance to guide further clinical treatment. In this paper, we analyze the performance of the current mainstream neural network models using the rectal tumor data set from the 7th Teddy Cup Data Mining Challenge B. Among them, ResUnet ++ achieves Dice value of 83.32% and IoU value of 70.06%, which is the best performance among mainstream models.