Automatic rib segmentation and labeling in computed tomography scans using a general framework for detection,...

Area of Science:

• Medical imaging analysis
• Radiology and medical physics
• Computer-aided diagnosis

Background:

• Accurate segmentation of the rib cage in CT scans is crucial for diagnosing various thoracic conditions.
• Manual segmentation is time-consuming and prone to inter-observer variability.
• Existing automated methods may lack comprehensive labeling and precise segmentation of the entire rib cage.

Purpose of the Study:

• To develop and validate a generalizable framework for automatic segmentation and labeling of the complete rib cage in 3D chest CT scans.
• To improve the efficiency and accuracy of rib cage analysis in radiological assessments.
• To provide a robust tool for quantitative analysis of thoracic structures.

Main Methods:

• A five-stage framework involving detection, primitive construction, classification, grouping, and segmentation was employed.
• 1D ridges were extracted, and line element primitives were generated from 3D CT data.
• A classifier distinguished rib primitives from background, followed by centerline formation, rib numbering, and seeded region growing for segmentation.

Main Results:

• The primitive classification achieved 97.5% accuracy (96.8% sensitivity, 97.8% specificity).
• Rib recognition after grouping reached 98.4% accuracy.
• The final segmentation demonstrated high accuracy for over 80% of ribs, with minor errors in remaining cases.

Conclusions:

• The proposed system effectively automates the segmentation and labeling of the complete rib cage in chest CT scans.
• The method demonstrates high accuracy and robustness, offering a valuable tool for clinical applications.
• Further refinement could address minor segmentation errors for enhanced clinical utility.