Relational Modeling for Robust and Efficient Pulmonary Lobe Segmentation in CT Scans.

Area of Science:

• Medical Imaging Analysis
• Artificial Intelligence in Healthcare
• Pulmonary Medicine

Background:

• Pulmonary lobe segmentation in CT scans is crucial for assessing lung diseases.
• Convolutional neural networks (CNNs) show promise but struggle with complex structural relationships.
• Accurate segmentation is vital for diseases like COVID-19 and COPD, where lung structure is altered.

Approach:

• Proposed RTSU-Net, a relational approach using a novel non-local neural network module.
• The module learns visual and geometric relationships for self-attention weights.
• Initial training on COPDGene dataset (5000 subjects), followed by transfer learning on COVID-19 suspects (470 subjects).

Key Points:

• RTSU-Net effectively captures inter-lobe shape and border relationships.
• The approach leverages structured relationships for improved segmentation accuracy.
• Demonstrated robustness in cases of severe lung infection due to COVID-19.

Conclusions:

• RTSU-Net outperforms existing methods in pulmonary lobe segmentation.
• The relational approach enhances accuracy, particularly in diseased lungs.
• This method holds potential for improved diagnosis and monitoring of pulmonary conditions.