Extraction of metastatic lymph nodes from MR images using two deformable model-based approaches

Area of Science:

• Medical Imaging
• Radiology
• Computational Anatomy

Background:

• Accurate delineation of metastatic cervical nodal lesions is crucial for nasopharyngeal carcinoma staging and treatment.
• Manual segmentation of these lesions can be time-consuming and subject to interobserver variability.

Purpose of the Study:

• To evaluate two deformable model-based approaches: region plus contour deformation (RPCD) and level sets.
• To compare the performance of these methods against manual tracing for extracting metastatic cervical nodes from MRI.

Main Methods:

• RPCD: Combines region deformation based on signal statistics with active contour deformation for boundary refinement.
• Level Sets: Utilizes an evolving contour guided by image gradients to identify nodal boundaries.

Main Results:

• RPCD achieved 82-84% averaged percentage matching, while level sets achieved 87-88%.
• Both deformable methods demonstrated significantly lower interoperator variance than manual tracing.
• Successful extraction of 33 metastatic cervical nodes from 18 patients' pretreatment T2-weighted MRI scans.

Conclusions:

• RPCD and level sets are effective automated tools for delineating metastatic cervical nodal lesions.
• These methods can aid in evaluating metastatic nodal volume for treatment response assessment.
• They offer viable alternatives to manual segmentation in radiation treatment planning.