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

Yifan Zhong, Yuxiang Dai, S. Li, Kejia Zhu, Yongpeng Lin, Ran Yu, Lin Chen, Ye Ruan, Pengfei Yu, Lin Li, Wenxiong Li, Chuang-Long Xu, Zhitao Sun, Kenneth A. Weber, De-Wei Kong, Feng Yang, Wen-Ping Lin, Jiang Chen, Bolai Chen, Hongwei� Jiang, Yingjie Zhou, Bo Sheng, Yongjun Wang, Yingzhong Tian, Yue-Li SunFrontiers in Bioengineering and Biotechnology 2024 · DOI

This paper introduces PDB-SLIC, a novel automated algorithm designed to measure cervical spine sagittal balance—a critical factor in diagnosing and treating neck-related degenerative diseases—directly from MRI scans. By combining multi-scale image processing with advanced clustering, the researchers achieved segmentation accuracy that rivals deep learning models while bypassing the need for massive, manually annotated training datasets.

TL;DR

Researchers developed an automated tool called PDB-SLIC that uses "superpixel" segmentation to measure cervical spine alignment in MRIs with over 95% correlation to expert surgeons. This unsupervised approach matches the performance of heavy-duty deep learning models like U-Net but is more efficient and robust across different hospital settings.

What Problem Does This Paper Address?

As our global population ages, cervical spondylosis (wear and tear of the spinal disks) is becoming a major health burden. To treat this effectively, surgeons need to assess "sagittal balance"—the alignment and curvature of the neck.

Currently, this assessment relies on manual measurements by specialists. This process is slow, subjective, and prone to human error. While deep learning (CNNs) can automate this, these models require thousands of "gold standard" images manually labeled by doctors to train effectively. Creating these datasets is a massive bottleneck. The authors sought a way to achieve high-precision automation without the "annotation tax."

Methods

The team’s solution is the Pyramid DBSCAN Simple Linear Iterative Cluster (PDB-SLIC). The workflow breaks down into three main phases:

  1. Preprocessing: They use CLAHE (Contrast Limited Adaptive Histogram Equalization) to sharpen the image contrast and Bilateral Filtering to remove noise while keeping the edges of the vertebrae crisp.
  2. Segmentation (The "PDB" part):
    • Pyramid: The algorithm looks at the image at multiple resolutions. This allows it to understand the "global" structure of the spine while still capturing "local" details.
  3. SLIC: It groups pixels into "superpixels"—small, uniform blobs that share similar colors and textures.
  4. DBSCAN: This is a density-based clustering algorithm used to merge those superpixels into distinct vertebral bodies.
  5. Measurement: Once the vertebrae are identified, the system automatically calculates the centroids and fits "minimum bounding rectangles" to the bones to determine clinical parameters like cervical curvature and the C7 slope.

Datasets

The study is impressively large-scale, utilizing a retrospective dataset of 4,258 T2-weighted mid-sagittal MR images.

  • Source: Data was collected from 10 different hospitals across China.
  • Diversity: The dataset included patients aged 20 to 90 and utilized scanners from major manufacturers (GE, Philips, Siemens) at both 1.5T and 3.0T field strengths.
  • Benchmarking: The PDB-SLIC results were compared against traditional superpixel methods (SLIC, VBSeg) and state-of-the-art deep learning networks (FCN, U-Net, DeepLabV3).

Key Insights and Ideas

  • The Unsupervised Advantage: The most striking insight is that PDB-SLIC performed nearly identically to U-Net (a gold-standard supervised deep learning model). The difference in the Jaccard index (a measure of overlap) was a mere 0.2%. This proves that clever algorithmic clustering can be just as effective as data-hungry neural networks for specific medical tasks.
  • Overcoming the "Extreme Value" Problem: Traditional superpixel methods often fail when a bone has very bright or very dark spots inside it. By using DBSCAN for merging, PDB-SLIC is much more robust to these internal intensity variations.
  • Clinical Validation: The study confirmed a clear trend: cervical curvature significantly increases with age. By providing an automated way to measure this, the tool helps establish "normative standards" that can tell a surgeon if a patient's alignment is typical for their age or a sign of pathology.

Key Results

  • Segmentation Accuracy: PDB-SLIC achieved a Jaccard index of 77.8% and a Precision of 80.0%, outperforming other unsupervised methods.
  • Clinical Correlation: The automated measurements showed a correlation coefficient exceeding 95% when compared to manual measurements by experienced spinal surgeons.
  • Error Rates: The mean absolute error for curvature was remarkably low (approx. 1.6° to 2.9° depending on the specific measurement method used), which is well within the acceptable margin for clinical decision-making.

Limitations and Future Work

The authors acknowledge a few hurdles for the future:

  • The "Supine" Factor: MRIs are taken while the patient is lying down (supine), which doesn't perfectly reflect the spine's alignment when standing and bearing weight. While MRI provides better soft-tissue detail than X-rays, the "true" balance is best seen standing.
  • Generalization: While 10 hospitals provide great diversity, expanding the dataset to include more global populations and different types of spinal deformities (like severe scoliosis) would further prove the tool's robustness.
  • Integration: The next logical step is integrating this algorithm into real-time surgical planning software to give surgeons instant feedback during preoperative assessments.

Summary updated

Abstract

<b>Introduction:</b> Magnetic Resonance Imaging (MRI) is essential in diagnosing cervical spondylosis, providing detailed visualization of osseous and soft tissue structures in the cervical spine. However, manual measurements hinder the assessment of cervical spine sagittal balance, leading to time-consuming and error-prone processes. This study presents the Pyramid DBSCAN Simple Linear Iterative Cluster (PDB-SLIC), an automated segmentation algorithm for vertebral bodies in T2-weighted MR images, aiming to streamline sagittal balance assessment for spinal surgeons. <b>Method:</b> PDB-SLIC combines the SLIC superpixel segmentation algorithm with DBSCAN clustering and underwent rigorous testing using an extensive dataset of T2-weighted mid-sagittal MR images from 4,258 patients across ten hospitals in China. The efficacy of PDB-SLIC was compared against other algorithms and networks in terms of superpixel segmentation quality and vertebral body segmentation accuracy. Validation included a comparative analysis of manual and automated measurements of cervical sagittal parameters and scrutiny of PDB-SLIC's measurement stability across diverse hospital settings and MR scanning machines. <b>Result:</b> PDB-SLIC outperforms other algorithms in vertebral body segmentation quality, with high accuracy, recall, and Jaccard index. Minimal error deviation was observed compared to manual measurements, with correlation coefficients exceeding 95%. PDB-SLIC demonstrated commendable performance in processing cervical spine T2-weighted MR images from various hospital settings, MRI machines, and patient demographics. <b>Discussion:</b> The PDB-SLIC algorithm emerges as an accurate, objective, and efficient tool for evaluating cervical spine sagittal balance, providing valuable assistance to spinal surgeons in preoperative assessment, surgical strategy formulation, and prognostic inference. Additionally, it facilitates comprehensive measurement of sagittal balance parameters across diverse patient cohorts, contributing to the establishment of normative standards for cervical spine MR imaging.
BibTeX
@inproceedings{zhong2024sagittal,
  author = {Yifan Zhong and Yuxiang Dai and S. Li and Kejia Zhu and Yongpeng Lin and Ran Yu and Lin Chen and Ye Ruan and Pengfei Yu and Lin Li and Wenxiong Li and Chuang-Long Xu and Zhitao Sun and Kenneth A. Weber and De-Wei Kong and Feng Yang and Wen-Ping Lin and Jiang Chen and Bolai Chen and Hongwei� Jiang and Yingjie Zhou and Bo Sheng and Yongjun Wang and Yingzhong Tian and Yue-Li Sun},
  title = {{Sagittal balance parameters measurement on cervical spine MR images based on superpixel segmentation}},
  booktitle = {Frontiers in Bioengineering and Biotechnology},
  year = {2024},
  doi = {10.3389/fbioe.2024.1337808},
}
Figure