Vertebral body segmentation in wide range clinical routine spine MRI data

Area of Science:

• Medical Imaging
• Computational Anatomy
• Radiology

Background:

• Vertebral body segmentation in MR spine imaging is crucial for clinical diagnosis.
• Existing methods are often slow or limited to specific sequences, hindering clinical application.
• Challenges include low image quality, high anisotropy, and diverse spine pathologies.

Purpose of the Study:

• To develop a rapid, precise, and broadly applicable 3D vertebral body segmentation method for clinical MR spine imaging.
• To minimize user interaction and computation time for routine clinical use.
• To ensure robustness across various MR imaging sequences and patient conditions.

Main Methods:

• The method employs intensity correction for bias field artifacts.
• It utilizes a minimal user-assisted initialization.
• A hybrid level-set segmentation is guided by appearance-based vertebral body probability maps.

Main Results:

• The method was evaluated on 63 datasets from 48 subjects, representing the largest reference set for clinical MR spine imaging.
• Achieved a Dice coefficient of 86.0%.
• Demonstrated a mean Euclidean surface distance error of 1.59 ± 0.24 mm and a Hausdorff distance of 6.86 mm.

Conclusions:

• The segmentation approach shows robustness across diverse MR imaging data.
• This adaptability is vital for clinical utility and reliable diagnosis.
• The method offers a significant improvement for routine clinical MR spine imaging analysis.