TY - GEN
T1 - Automatic recognition and tracking of liver blood vessels in ultrasound image using deep neural networks
AU - Zhao, Yue
AU - Wang, Yuanzheng
AU - Yu, Yuan
AU - Yang, Feng
AU - Shen, Yi
N1 - Publisher Copyright:
© 2020 IEEE.
PY - 2020/12/6
Y1 - 2020/12/6
N2 - Medical ultrasound devices are widely used in clinic because of its convenience, rapid and non-invasive. But ultrasound (US) images have the characteristics of large speckle noise, unclear target and low brightness. Since the deep learning theory has been developed, the accuracy of the tasks in the field of image has been greatly improved. In this paper, a deep neural network structure is established to automatic detect and track the liver vessel targets. Firstly, the dataset is augmented and preprocessed using histogram equalization. Secondly, the RetinaNet is implemented to extract the region of interest (ROI) in the US image. Then, the U-net is used to extract the features of the ROI, and deconvolution is implemented to restore the feature matrix to the size of the original image, which realize the automatic segmentation of blood vessels. Finally, the LSTM network is used to predict the information of vessels in the subsequent image. Experimental results show that the proposed algorithm is fast and robust. The accuracy of the ROI detection is 98.9%. The average error of the distance of the center point of the target is less than 1 mm.
AB - Medical ultrasound devices are widely used in clinic because of its convenience, rapid and non-invasive. But ultrasound (US) images have the characteristics of large speckle noise, unclear target and low brightness. Since the deep learning theory has been developed, the accuracy of the tasks in the field of image has been greatly improved. In this paper, a deep neural network structure is established to automatic detect and track the liver vessel targets. Firstly, the dataset is augmented and preprocessed using histogram equalization. Secondly, the RetinaNet is implemented to extract the region of interest (ROI) in the US image. Then, the U-net is used to extract the features of the ROI, and deconvolution is implemented to restore the feature matrix to the size of the original image, which realize the automatic segmentation of blood vessels. Finally, the LSTM network is used to predict the information of vessels in the subsequent image. Experimental results show that the proposed algorithm is fast and robust. The accuracy of the ROI detection is 98.9%. The average error of the distance of the center point of the target is less than 1 mm.
KW - US image
KW - automatic detection and segmentation
KW - deep neural network
KW - target prediction and tracking
UR - https://www.scopus.com/pages/publications/85100270373
U2 - 10.1109/ICSP48669.2020.9320944
DO - 10.1109/ICSP48669.2020.9320944
M3 - 会议稿件
AN - SCOPUS:85100270373
T3 - International Conference on Signal Processing Proceedings, ICSP
SP - 499
EP - 504
BT - ICSP 2020 - 2020 IEEE 15th International Conference on Signal Processing Proceedings
A2 - Baozong, Yuan
A2 - Qiuqi, Ruan
A2 - Yao, Zhao
A2 - Gaoyun, An
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 15th IEEE International Conference on Signal Processing, ICSP 2020
Y2 - 6 December 2020 through 9 December 2020
ER -