Ca-Segresnet: A Context-Aware Segmentation Residual Network for Automatic Segmentation of Identified Vertebral Bones from Ct Images

17 Pages Posted: 30 Nov 2023

See all articles by Zhongqi Zhu

Zhongqi Zhu

East China Normal University (ECNU)

Xiaolong Gao

Fujin Traditional Chinese Medicine Hospital

Yinghao Li

East China Normal University (ECNU)

Liguo Hao

Qiqihar Medical University

Guang Yang

East China Normal University (ECNU) - Shanghai Key Laboratory of Magnetic Resonance

Hongzhi Wang

East China Normal University (ECNU)

Abstract

Accurate segmentation of vertebral bones in computer tomography (CT) spine images is critical for the diagnosis and treatment of spinal diseases. In this paper, we propose an improved network called CA-SegResNet, which utilizes residual convolutional neural networks to extract image features. CA-SegResNet combines the feature maps outputted by each layer of the encoder with the inputs of each layer of the decoder. Additionally, it introduces a Three-Dimensional Coordinate Attention module to capture inter-channel relationships, direction, and position information, allowing for the establishment of long-range dependencies in different spatial directions for the localization and segmentation of identified vertebral bones. In the segmentation task of cervical vertebrae (C7) and thoracic vertebrae (T12) for each of the 105 cases, CA-SegResNet achieved a localization accuracy rate of 100% for identified vertebrae on the test dataset. The average Dice coefficients for the segmentation results were 93.45% and 91.89%, with average Hausdorff distances of 7mm and 8mm, respectively. Compared to UNet, CA-SegResNet shows an increase of 0.0145 and 0.0463 in average Dice coefficients, and a decrease in average Hausdorff distances of 176mm and 388mm for C7 and T12, respectively. The results suggest that the proposed model can accurately segment vertebral spines on CT images and identify them based on their localization. Moreover, the study can potentially save up to 90% of the workload of manual annotation compared with segmenting the entire spine.

Note:
Funding declaration: Financial support from Shanghai Domestic Science and Technology Cooperation Project (21015801000) is gratefully acknowledged.

Conflict of Interests: The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Ethical Approval: This study is a retrospective study and has been exempted by the local institutional ethics review board. Before the scans, patients were informed about the scanning procedure, obtained their consent, and signed informed consent forms.

Keywords: 【Keywords】Spine segmentation, CT images, CA-SegResNet, Deep learning, Coordinate Attention

Suggested Citation

Zhu, Zhongqi and Gao, Xiaolong and Li, Yinghao and Hao, Liguo and Yang, Guang and Wang, Hongzhi, Ca-Segresnet: A Context-Aware Segmentation Residual Network for Automatic Segmentation of Identified Vertebral Bones from Ct Images. Available at SSRN: https://ssrn.com/abstract=4629021 or http://dx.doi.org/10.2139/ssrn.4629021

Zhongqi Zhu

East China Normal University (ECNU) ( email )

Xiaolong Gao

Fujin Traditional Chinese Medicine Hospital ( email )

Yinghao Li

East China Normal University (ECNU) ( email )

Liguo Hao

Qiqihar Medical University ( email )

Qiqihar
China

Guang Yang

East China Normal University (ECNU) - Shanghai Key Laboratory of Magnetic Resonance ( email )

China

Hongzhi Wang (Contact Author)

East China Normal University (ECNU) ( email )

Do you have a job opening that you would like to promote on SSRN?

Paper statistics

Downloads
94
Abstract Views
563
Rank
608,612
PlumX Metrics