Reconstruction of 3D lumbar spine models from incomplete segmentations using landmark detection

Area of Science:

• Biomedical Engineering
• Medical Imaging
• Computational Anatomy

Background:

• Patient-specific 3D spine models are crucial for treatment planning and biomechanical research.
• Limitations in imaging and segmentation often result in incomplete 3D spine models, hindering applications.
• Reconstructing complete lumbar spine models from incomplete data remains a challenge.

Purpose of the Study:

• To develop a novel, accurate, and time-efficient method for reconstructing complete 3D lumbar spine models from incomplete vertebral bodies.
• To address challenges posed by missing spinal structures in medical images and segmentations.
• To provide a foundation for advanced spine diagnostics and treatment planning.

Main Methods:

• Utilized an affine transformation to align artificial vertebra models with incomplete patient-specific vertebrae from segmented MRI.
• Employed automatic detection of vertebra landmarks on endplates to derive the transformation matrix.
• Registered the entire lumbar spine (L1-L5) using the developed method.

Main Results:

• Achieved high registration accuracy with an average point-to-model distance of 1.95 mm.
• Demonstrated a mean absolute error of 3.4° for functional spine unit angles, preserving morphological properties.
• Completed the registration of the L1-L5 lumbar spine in an average of 0.14 seconds, highlighting time-efficiency.

Conclusions:

• The novel method enables fast and accurate reconstruction of complete 3D lumbar spine models from incomplete data.
• This technique has significant clinical relevance for spine diagnostics, surgical planning, and healthcare solutions.
• The approach enhances the utility of patient-specific spine models in research and clinical practice.