[01] Shijie Xing, School of Information Engineering, China University of Geosciences (Beijing), Beijing, China. [02] Qiaodan Zhang, Beijing Kwai Technology Co., Ltd, Beijing, China. [03] Mingying Zhang, School of Information Engineering, China University of Geosciences (Beijing), Beijing, China. [04] Mei Li, School of Information Engineering, China University of Geosciences (Beijing), Beijing, China.

Uterine adenomyoma is a common disease in women. Its symptoms and pain have seriously troubled the physical and mental health of contemporary women. Because of the advantages of no damage to the body and low cost, so ultrasound is usually used to obtain the medical image of adenomyoma in medicine. Uterine adenomyoma is different from human epidermal diseases. The boundary of the image obtained by ultrasound is not clear and the interference is large. Moreover, due to the influence of noise and artifact of ultrasound image, image segmentation is difficult, and the clinical diagnosis results are lack of reliability. In order to solve the above problems, this paper preprocesses the obtained ultrasound images, then constructs the ultrasound image data set, sets the network parameters, inputs the ultrasound images into the deeplab model network for training, and finally optimizes the edge details of the lesions through the hole convolution algorithm and the full connection conditional random field, and uses the average intersection ratio and pixel accuracy as the evaluation criteria for semantic segmentation. In order to ensure the accuracy of the test data and the superiority of the algorithm. And constantly optimize the network training to achieve the fine segmentation of the lesion area. The accuracy and feasibility of the model in medical image segmentation are demonstrated by experiments, which fills the blank of deep learning in this application field.

Adenomyoma, Ultrasound Image, Deeplab Model, Hole Convolution Algorithm, Deep Learning

[01] Kennedy, & E., J. (2005). Innovation: High-intensity focused ultrasound in the treatment of solid tumours. 5 (4), 321-327. [02] Ping Yang (2013). Types of gynecological diseases and their harm to women. [03] Xiao Lin & Xiaojia Qiu. (2005). Application of image analysis technology in medicine. (3). [04] Xianshan Zhou. (2009). Review of medical image processing technology. 025 (1), 34, 33. [05] Carneiro, G., Nascimento, J. C., & Freitas, A. (2011). Semi-supervised self-trainingmodel for the segmentationof the left ventricle of the heart from ultrasound data. Paper presented at the IEEE International Symposium on Biomedical Imaging: from Nano to Macro. [06] Looney, P., Stevenson, G. N., Nicolaides, K. H., Plasencia, W., Molloholli, M., Natsis, S., & Collins, S. L. (2017). Automatic 3D ultrasound segmentation of the first trimester placenta using deep learning. Paper presented at the IEEE International Symposium on Biomedical Imaging. [07] Zhang, Y., Ying, M. T. C., Lin, Y., Ahuja, A. T., & Chen, D. Z. (2016). Coarse-to-Fine Stacked Fully Convolutional Nets for lymph node segmentation in ultrasound images. Paper presented at the IEEE International Conference on Bioinformatics & Biomedicine. [08] Milletari, F., Ahmadi, S. A., Kroll, C., Plate, A., Rozanski, V., Maiostre, J., ... Understanding, I. (2017). Hough-CNN: Deep learning for segmentation of deep brain regions in MRI and ultrasound. 164 (NOV.), 92-102. [09] caojia Liang. (2019). Research on pavement state recognition algorithm based on deep semantic segmentation network. [10] Ronneberger, O., Fischer, P., & Brox, T. (2015). U-Net: Convolutional Networks for Biomedical Image Segmentation. [11] Rumelhart, D. E., Hinton, G. E., & Williams, R. J. J. N. (1986). Leaning internal representations by back-propagating errors. 323 (6088), 318-362. [12] Yuan, X., Yuxin, W., Jie, Y., Qian, C., Xueding, W., & Ultrasonics, C. P. L. J. (2019). Medical breast ultrasound image segmentation by machine learning. 91, 1-9. [13] Zhang, W., Li, R., Deng, H., Wang, L., Lin, W., Ji, S., & Shen, D. J. N. (2015). Deep convolutional neural networks for multi-modality isointense infant brain image segmentation. 108, 214-224. [14] Zhou, Y. T., & Chellappa, R. (1988). Computation of optical flow using a neural network. Paper presented at the Neural Networks, 1988., IEEE International Conference on. [15] Xiaoxiao Du. (2019). Research on Key Technologies of computer aided diagnosis of early esophageal cancer based on semantic segmentation. Zhengzhou University, Henan. [16] Shuping Jia, Liang Zhao, & hejuan Jiang. (2019). Application progress of high intensity focused ultrasound ablation in tumor treatment. Chinese medical device information, 025 (009), 48-50. [17] Ru Jiang, &Li jun Liao. (2010). Changes of estrogen receptor in endometrium myometrium interface of patients with adenomyosis. (21), 101-102.