Smoothing lung segmentation surfaces in three-dimensional X-ray CT images using anatomic guidance

Area of Science:

• Medical Imaging
• Computational Anatomy

Background:

• Automatic lung segmentation in computed tomography (CT) is crucial but faces challenges with inconsistent boundaries near the mediastinum.
• Existing methods relying on CT attenuation values produce irregular lung surfaces, impacting processes like image registration and atlas construction.

Purpose of the Study:

• To develop a fully automatic, three-dimensional method for smoothing the lung boundary, specifically addressing inconsistencies around the mediastinum.
• To improve the consistency and accuracy of lung segmentation in volumetric CT images.

Main Methods:

• A novel approach utilizing the segmented human airway tree to define and smooth the lung boundary.
• Operations are confined to a bounding box around the mediastinum, using airway tree structures distal to the main bronchi to delineate lung regions.
• A fast morphological closing operation with an ellipsoidal kernel is applied for surface smoothing.

Main Results:

• The smoothing method significantly reduced the curvedness of lung contours (from 4.91 to 0.13).
• Accuracy improved, with the root mean square difference between automated and manual borders decreasing from 1.3012 mm to 0.8691 mm.
• A high mean similarity index of 0.9958 (SD 0.0032) was achieved, indicating excellent overlap.

Conclusions:

• A novel, automatic scheme for smoothing lung contours near the mediastinum using airway tree anatomy has been developed.
• Validation demonstrates good agreement between automated smoothing and manual tracing, with improved accuracy and consistency.
• The airway tree-based boundary definition offers a robust basis for 3D lung surface smoothing in medical imaging.