Page 48 - AIH-2-1

Artificial Intelligence in Health

ORIGINAL RESEARCH ARTICLE
Enhancing spinal MRI segmentation with an

asymmetric U-Net architecture

2
2†
3
Longfei Zhou * , Xingyu Chen , Weihao Cheng , Zhanghao Qin ,
1†
4
5
2
6
Tianao Shen , Pingyu Cao , Zebo Huang , Xiangyu Wu , and Yiyao Zhang 7
1 Department of Biomedical, Industrial and Systems Engineering, College of Engineering and
Business, Gannon University, Erie, Philadelphia, United States of America
2 School of Artificial Intelligence and Computer Science, Jiangnan University, Wuxi, Jiangsu, China
3 School of Physics and Optoelectronic Engineering, Nanjing University of Information Science and
Technology, Nanjing, Jiangsu, China
4 School of Internet of Things Engineering, Jiangnan University, Wuxi, Jiangsu, China
5 School of Remote Sensing and Surveying Engineering, Nanjing University of Information Science
and Technology, Nanjing, Jiangsu, China
6 School of Mechanical Engineering, Jiangnan University, Wuxi, Jiangsu, China
7 College of Robot and Engineering, Guangzhou City University of Technology, Yinchuan, Ningxia,
China

† These authors contributed equally
to this work. Abstract
*Corresponding author:
Longfei Zhou Spinal diseases are among the most prevalent health issues in modern society,
(zhou009@gannon.edu) significantly impacting patients’ quality of life. Diagnosing conditions such as disc
herniation and spinal deformity requires advanced medical imaging techniques,
Citation: Zhou L, Chen X,
Cheng W, et al. Enhancing including X-rays, magnetic resonance imaging (MRI), computed tomography, and
spinal MRI segmentation with an nuclear magnetic resonance. Spine MRI is particularly crucial due to its ability to
asymmetric U-Net architecture. provide high-resolution images of soft tissues, essential for accurate diagnosis.
Artif Intell Health. 2025;2(1):42-52.
doi: 10.36922/aih.3889 However, the manual segmentation of spine MRI images is labor-intensive and
inadequate for large-scale quantitative analysis. Thus, developing automated
Received: June 7, 2024
spinal MRI segmentation methods is critical to alleviating doctors’ workload and
1st revised: July 5, 2024 enhancing diagnostic efficiency. In this study, we propose a novel asymmetric
2nd revised: July 12, 2024 U-Net architecture designed to improve the precision of reconstructing complex
structures and details by increasing the depth of the upsampling side. The
Accepted: August 1, 2024
model incorporates adjacent-scale skip connections to control parameters while
Published Online: October 21, maintaining high segmentation accuracy. In addition, residual connections on the
2024
upsampling side prevent gradient vanishing, thereby enhancing the network’s
Copyright: © 2024 Author(s). feature learning and representation capabilities. Experimental results indicate
This is an Open-Access article that this method significantly reduces training time and increases model accuracy
distributed under the terms of the
Creative Commons Attribution compared to traditional approaches, marking a substantial advancement in
License, permitting distribution, automated spinal MRI segmentation. This innovative approach holds promise for
and reproduction in any medium, improving clinical outcomes and optimizing the workflow in medical imaging
provided the original work is
properly cited. departments.
Publisher’s Note: AccScience
Publishing remains neutral with Keywords: Spinal magnetic resonance imaging; Automated segmentation; Asymmetric
regard to jurisdictional claims in
published maps and institutional U-Net; Medical imaging; Deep learning
affiliations.

Volume 2 Issue 1 (2025) 42 doi: 10.36922/aih.3889