BranchLabelNet: Anatomical Human Airway Labeling Approach using a Dividing-and-Grouping Multi-Label Classification

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Pulmonology

Background:

• Accurate anatomical airway labeling is vital for diagnosing pulmonary diseases like asthma and COPD.
• Current methods struggle with the complex, fractal nature of airway branching.

Purpose of the Study:

• To develop an innovative airway labeling methodology, BranchLabelNet, that accurately identifies airway structures.
• To improve the diagnosis and treatment of pulmonary ailments through precise anatomical labeling.

Main Methods:

• Extracted branch-related parameters (position vectors, generation levels, lengths, areas, perimeters) from 1000 chest CT images.
• Utilized an n-ary tree structure to manage complex airway relationships.
• Employed a divide-and-group deep learning approach for multi-label classification and the Tomek Links algorithm to address class imbalance.

Main Results:

• Achieved an average classification accuracy of 95.94% across fivefold cross-validation.
• Demonstrated robust branch designations for anatomical airways.
• Successfully managed complex branch data using an n-ary tree structure.

Conclusions:

• BranchLabelNet offers a reliable and accurate solution for anatomical airway labeling.
• The methodology is adaptable for general multi-label classification in biomedical systems.
• This AI-driven approach enhances diagnostic capabilities for pulmonary conditions.