A Fissure-Aided Registration Approach for Automatic Pulmonary Lobe Segmentation Using Deep Learning.

Area of Science:

• Medical Imaging
• Computational Anatomy
• Radiology

Background:

• Pulmonary lobe segmentation is vital for clinical applications like lesion localization and surgical planning.
• Automatic segmentation faces challenges due to incomplete fissures and lung disease-induced morphological variations.

Purpose of the Study:

• To develop a robust and accurate learning-based method for pulmonary lobe segmentation.
• To incorporate local fissure information, whole lung context, and prior anatomical knowledge.

Main Methods:

• A learning-based approach using registered pulmonary atlases and detected fissures.
• Deformation mapping from an annotated atlas to segment lobes.
• Evaluation on a custom dataset of CT scans from patients with COPD.

Main Results:

• Achieved high average Dice coefficients: 0.95 (right upper), 0.90 (right middle), 0.97 (right lower), 0.97 (left upper), and 0.97 (left lower).
• Demonstrated comparable accuracy to previous methods.
• Showed improved robustness in segmenting lungs with specific morphological changes.

Conclusions:

• The proposed method offers a robust and accurate solution for pulmonary lobe segmentation.
• It effectively handles variations caused by lung diseases like COPD.
• This technique has significant potential for improving clinical assessment and surgical planning.