Uterine magnetic resonance image segmentation based on deep learning

Abstract

Endometrial diseases have always been a common disease among women worldwide. For the prevention of endometrial diseases and the treatment after illness, there is an urgent need to solve the problem of positioning the uterus, which is also one of the difficult problems that plague clinicians. Research in recent years has shown that neural networks based on deep learning are an important tool in the medical field. This article discusses the uterus magnetic resonance image segmentation method based on deep learning. First, we completed the preprocessing of the data based on the hessian matrix, and expanded the data, then inputted the data into the DenseUNet network. In this article, our method obtained 87.60% Dice Similariy Coefficient(DSC), 86.57% precision, 88.11% sensitivity and 99.75% specificity in 13 different test sets. Finally, our method was compared to UNet and UNet++ networks, and achieved better performance. Our method effectively solved the problems of uterus magnetic resonance image automatic segmentation. It can be used in the pre-operation planning of stem cell surgery to repair the endometrium.