Deep learning-based bronchial tree-guided semi-automatic segmentation of pulmonary segments in computed tomography...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Pulmonology

Background:

• Pulmonary segments offer greater precision than lobes for lung cancer localization in CT images.
• Advances in precision therapy necessitate accurate identification and visualization of pulmonary segments for lung cancer treatment.

Purpose of the Study:

• To integrate multiple deep learning models for accurate pulmonary segment segmentation in CT images.
• To utilize a bronchial tree (BT)-based approach for enhanced segmentation accuracy.

Main Methods:

• A five-step method involving lung, lobe, and airway segmentation using U-Net, V-Net, and BronchiNet models.
• Bronchial tree branch labeling based on anatomical position.
• Pulmonary segment segmentation based on voxel distance to labeled BT branches.

Main Results:

• High Dice Similarity Coefficients (DSC) for lung (0.98±0.03) and lobe (0.94±0.06) segmentation.
• Accurate airway segmentation with an average tree length of 1,902.8±502.1 mm and 8.5±1.3 generations.
• Pulmonary segment segmentation achieved a DSC of 0.73±0.11 and a mean surface distance of 6.1±2.9 mm.

Conclusions:

• Demonstrated feasibility of combining deep learning models for auxiliary pulmonary segment segmentation on CT images.
• The BT-based method shows potential for semi-automatic segmentation of pulmonary segments.