Fast interactive segmentation of the pulmonary lobes from thoracic computed tomography data

Area of Science:

• Medical Imaging
• Computer-Aided Diagnosis
• Pulmonology

Background:

• Automated lung lobe segmentation methods often fail in cases with incomplete fissures or significant pathology.
• Accurate lung lobe segmentation is crucial for clinical diagnosis and analysis.

Purpose of the Study:

• To present a fast and intuitive method for interactively correcting or creating lung lobe segmentations.
• To evaluate the method's performance in terms of speed, accuracy, and user interaction.

Main Methods:

• A mesh-based approach using principal component analysis of 3D lobar boundary markers.
• Interactive modification of the mesh by drawing on 2D slices, with immediate 3D adaptation.
• Evaluation using the LObe and lung analysis 2011 (LOLA11) challenge dataset.

Main Results:

• Interactive correction significantly improved segmentation accuracy, reducing average distance to reference from 2.68 mm to under 0.9 mm.
• Segmentation creation from scratch achieved an average distance of 0.77 mm with minimal interaction.
• Correction and creation tasks were completed quickly, with few interactions per case.

Conclusions:

• The interactive method is feasible for rapid and accurate lung lobe segmentation, even in challenging cases with pathology.
• The approach offers a valuable tool for both correcting automated segmentations and creating them de novo.
• Its independence from image features allows successful application in patients with severe pathologies.