3D Segmentation of Uterine Fibroids Based on Deep Supervision and an Attention Gate

Zhiwei Liu ¹ Chengnv Sun ² Chengwei Li ² Fajin Lv ^2*

• ¹ Ganzhou Cancer Hospital, Jiang xi, China
• ² State Key Laboratory of Ultrasound in Medicine and Engineering, Chongqing Medical University, Chongqing, China

The segmentation of uterine fibroids is very important for the treatment of patients. However, uterine fibroids are small and have a low contrast with surrounding tissue, making this task very challenging. To solve these problems, this paper proposes a 3D VNet automatic segmentation method for uterine fibroids based on deep supervision and attention gates and accurately segments uterine fibroids in magnetic resonance images using convolutional information. A deep supervision mechanism is introduced in the hidden layer. We introduce an attention gate during upsampling to increase attention to regions of interest. The network structure consists of VNet, a deep supervision module and an attention gate module. A dataset with 245 cases of uterine fibroids was divided into a training set, validation set and testing set at a ratio of 6:2:2. A total of 147 patient T2-weighted imaging (T2WI) magnetic resonance (MR) images were used for training, 49 were used for validation, and the algorithm was tested with 49 patient MR images. The experimental results showed that the proposed method achieved satisfactory segmentation results. The Dice similarity coefficient (DSC), intersection over union (IOU), sensitivity, precision and Hausdorff distance (HD) were 0.878, 0.784, 0.879, 0.885 and 11.180 mm, respectively. This proves that the proposed method can improve the automatic segmentation accuracy of magnetic resonance image (MRI) data of uterine fibroids.