AUTHOR=Zhong Yi-Fan , Dai Yu-Xiang , Li Shi-Pian , Zhu Ke-Jia , Lin Yong-Peng , Ran Yu , Chen Lin , Ruan Ye , Yu Peng-Fei , Li Lin , Li Wen-Xiong , Xu Chuang-Long , Sun Zhi-Tao , Weber Kenneth A. , Kong De-Wei , Yang Feng , Lin Wen-Ping , Chen Jiang , Chen Bo-Lai , Jiang Hong , Zhou Ying-Jie , Sheng Bo , Wang Yong-Jun , Tian Ying-Zhong , Sun Yue-Li 

TITLE=Sagittal balance parameters measurement on cervical spine MR images based on superpixel segmentation

JOURNAL=Frontiers in Bioengineering and Biotechnology

VOLUME=Volume 12 - 2024

YEAR=2024

URL=https://www.frontiersin.org/journals/bioengineering-and-biotechnology/articles/10.3389/fbioe.2024.1337808

DOI=10.3389/fbioe.2024.1337808

ISSN=2296-4185

ABSTRACT=Magnetic Resonance Imaging (MRI) plays a critical role in diagnosing cervical spondylosis, offering clear visualization of both osseous and soft tissue structures within the cervical spine. However, the evaluation of the cervical spine sagittal balance is impeded by the manual nature of measurements, leading to time-consuming and error-prone processes. Our study introduces the Pyramid DBSCAN Simple Linear Iterative Cluster (PDB-SLIC), an innovative automatic segmentation algorithm for vertebral bodies in T2-weighted MR images, with the aim of providing spinal surgeons with a streamlined tool for sagittal balance assessment. Developed by amalgamating the SLIC superpixel segmentation algorithm with DBSCAN clustering, PDB-SLIC underwent rigorous testing utilizing an extensive dataset comprising T2-weighted mid-sagittal MR images from 4,258 patients across ten hospitals in China. The efficacy of PDB-SLIC was gauged against the SLIC algorithm, the VBseg algorithm, Fully Convoltutional Networks (FCN)  network, DeeplabV3 network, and U-Net network in terms of superpixel segmentation quality and vertebral body segmentation accuracy. Subsequent validation encompassed a comparative analysis between manual and automated measurements of cervical sagittal parameters, alongside  scrutiny of PDB-SLIC’s measurement stability across MR images sourced from diverse hospitals and MR scanning machines. Results underscored PDB-SLIC’s superiority over SLIC and VBseg algorithms in vertebral body segmentation quality, closely rivaling the performance of deep learning networks with elevated average accuracy, recall, and Jaccard index. Moreover, PDB-SLIC exhibited minimal error deviation compared to manual measurements in cervical curvature measurement algorithm error verification, achieving correlation coefficients exceeding 95%. In the realm of generalization verification, PDB-SLIC demonstrated commendable performance in processing cervical spine T2-weighted MR images from heterogeneous hospital settings, MRI machines, and patient demographics. In conclusion, the PDB-SLIC algorithm emerges as an accurate, objective, and expeditious instrument for evaluating the sagittal balance of the cervical spine, furnishing indispensable assistance to spinal surgeons in preoperative assessment, surgical strategy delineation, and prognostic inference. Furthermore, this technological advancement facilitated comprehensive measurement of sagittal balance parameters across sizable cohorts, thereby facilitatingthe establishment of standard normative standards for cervical spine MR imaging.