Computer-aided lymph node segmentation in volumetric CT data.

Area of Science:

• Medical Imaging
• Radiology
• Computer-Aided Diagnosis

Background:

• Accurate lymph node segmentation in CT images is crucial for clinical applications such as biopsy planning and treatment response assessment.
• Challenges include partial volume effects, similar intensity profiles of neighboring structures, and inhomogeneous lymph node density.

Purpose of the Study:

• To develop and validate a computer-aided method for the 3D segmentation of lymph nodes in CT images.
• To facilitate applications like biopsy planning, image-guided radiation treatment, and assessment of response to therapy.

Main Methods:

• An optimal surface finding (OSF) based method was developed using graph generation and optimization.
• An interactive computer-aided segmentation refinement algorithm was incorporated to correct errors.
• Validation involved 111 lymph nodes from 35 CT scans across various anatomical regions.

Main Results:

• The method achieved mean Dice coefficients of 0.847 ± 0.061 (Set I), 0.836 ± 0.058 (Set II), and 0.809 ± 0.070 (Set III).
• Average signed surface distance errors were 0.023 ± 0.171 mm (Set I), 0.394 ± 0.189 mm (Set II), and 0.001 ± 0.146 mm (Set II).
• Initial segmentation required refinement in 36% of cases, with refinement taking approximately 10 seconds per lymph node.

Conclusions:

• The developed computer-aided method effectively addresses challenges in lymph node segmentation from CT images.
• The OSF approach provides accurate initial segmentations, and the refinement framework efficiently handles segmentation errors.
• The entire process, including refinement, typically takes less than one minute per lymph node, enhancing clinical workflow.