Vertebrae, IVD and spinal canal boundary extraction on MRI, utilizing CT-trained active shape models

Area of Science:

• Medical Imaging
• Computational Anatomy
• Biomedical Engineering

Background:

• Spinal cord pathologies often stem from displacements in vertebrae, intervertebral discs (IVD), and the spinal canal (SC).
• Accurate localization and boundary extraction of these spinal structures are crucial for diagnosing and assessing pathological conditions.
• Current methods may require paired imaging or large datasets, limiting their applicability.

Purpose of the Study:

• To develop a computational method for precise boundary extraction of vertebrae, IVD, and SC in magnetic resonance images (MRI).
• To utilize cross-modality shape priors (CT to MRI) for improved segmentation accuracy.
• To create a method that is independent of paired CT/MR scans and large training datasets.

Main Methods:

• Employing vertebrae shape priors derived from computed tomography (CT) images to guide segmentation in MRI.
• Utilizing active shape models (ASMs) trained on CT data to extract boundary information in MRI.
• Leveraging the similarity in edge locations between CT and MR images for cross-modality transfer.

Main Results:

• The proposed method achieves segmentation results comparable to state-of-the-art techniques on MR image datasets.
• CT-derived shape priors demonstrate superior accuracy for boundary extraction compared to MRI-derived priors.
• The bimodal strategy effectively transfers shape information across imaging modalities.

Conclusions:

• The developed method offers a robust alternative to existing bimodal approaches by not requiring paired CT/MR images.
• It overcomes limitations of deep learning methods by not needing large training datasets.
• The approach requires minimal user intervention, enhancing its practical utility in clinical settings.