Lung vessel connectivity map as anatomical prior knowledge for deep learning-based lung lobe segmentation.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Automated segmentation of lung lobes in chest CT scans is crucial for clinical analysis.
• Deep learning (DL) methods offer potential but can struggle with complex anatomical variations.
• Incorporating prior anatomical knowledge may improve DL model performance and robustness.

Purpose of the Study:

• To investigate the benefits of integrating prior anatomical knowledge into DL for lung lobe segmentation.
• To develop and evaluate a DL approach using lung vessel connectivity (LVC) maps.
• To assess the impact of LVC on segmentation accuracy, generalization, and robustness.

Main Methods:

• An automated DL-based approach using LVC maps was developed.
• Three neural network architectures (U-Net, multitasking U-Net, cascade U-Net) within nnU-Net were explored.
• The method was tested on chest CT scans, including cases with COVID-19.

Main Results:

• Inclusion of LVC information improved segmentation accuracy, particularly in challenging boundary regions of expiration CT scans.
• LVC demonstrated potential to enhance the DL model's generalization capabilities.
• The method showed robustness and applicability in segmenting lung lobes in COVID-19 cases.

Conclusions:

• Integrating prior anatomical information like LVC shows promise for improving DL-based lung lobe segmentation.
• Improvements were most notable in boundary regions, but limitations exist.
• Further research is needed to fully explore the practical applicability of this approach.