TY - GEN
T1 - TransBridge: a lightweight transformer for left ventricle segmentation in echocardiography
AU - Deng, Kaizhong
AU - Meng, Yanda
AU - Gao, Dongxu
AU - Bridge, Joshua
AU - Shen, Yaochun
AU - Lip, Gregory Y. H.
AU - Zhao, Yitian
AU - Zheng, Yalin
PY - 2021/9/21
Y1 - 2021/9/21
N2 - Echocardiography is an essential diagnostic method to assess cardiac functions. However, manually labelling the left ventricle region on echocardiography images is time-consuming and subject to observer bias. Therefore, it is vital to develop a high-performance and efficient automatic assessment tool. Inspired by the success of the transformer structure in vision tasks, we develop a lightweight model named ‘TransBridge’ for segmentation tasks. This hybrid framework combines a convolutional neural network (CNN) encoder-decoder structure and a transformer structure. The transformer layers bridge the CNN encoder and decoder to fuse the multi-level features extracted by the CNN encoder, to build global and inter-level dependencies. A new patch embedding layer has been implemented using the dense patch division method and shuffled group convolution to reduce the excessive parameter number in the embedding layer and the size of the token sequence. The model is evaluated on the EchoNet-Dynamic dataset for the left ventricle segmentation task. The experimental results show that the total number of parameters is reduced by 78.7% compared to CoTr and the Dice coefficient reaches 91.4%, proving the structure’s effectiveness.
AB - Echocardiography is an essential diagnostic method to assess cardiac functions. However, manually labelling the left ventricle region on echocardiography images is time-consuming and subject to observer bias. Therefore, it is vital to develop a high-performance and efficient automatic assessment tool. Inspired by the success of the transformer structure in vision tasks, we develop a lightweight model named ‘TransBridge’ for segmentation tasks. This hybrid framework combines a convolutional neural network (CNN) encoder-decoder structure and a transformer structure. The transformer layers bridge the CNN encoder and decoder to fuse the multi-level features extracted by the CNN encoder, to build global and inter-level dependencies. A new patch embedding layer has been implemented using the dense patch division method and shuffled group convolution to reduce the excessive parameter number in the embedding layer and the size of the token sequence. The model is evaluated on the EchoNet-Dynamic dataset for the left ventricle segmentation task. The experimental results show that the total number of parameters is reduced by 78.7% compared to CoTr and the Dice coefficient reaches 91.4%, proving the structure’s effectiveness.
KW - echocardiography
KW - left ventricle segmentation
KW - lightweight transformer model
KW - parameter efficiency
UR - https://dblp.org/db/conf/miccai/asmus2021.html#DengMGBSLZZ21
UR - https://dblp.org/db/conf/miccai/2021asmus
U2 - 10.1007/978-3-030-87583-1_7
DO - 10.1007/978-3-030-87583-1_7
M3 - Conference contribution
SN - 9783030875824
T3 - Lecture Notes in Computer Science (part of the Image Processing, Computer Vision, Pattern Recognition, and Graphics sub series)
SP - 63
EP - 72
BT - Simplifying Medical Ultrasound. ASMUS 2021
A2 - Noble, J. Alison
A2 - Aylward, Stephen
A2 - Grimwood, Alexander
A2 - Min, Zhe
A2 - Lee, Su-Lin
A2 - Hu, Yipeng
PB - Springer
T2 - ASMUS 2021: Second International Workshop on Advances in Simplifying Medical Ultrasound
Y2 - 27 September 2021 through 27 September 2021
ER -