Fuzzy Attention Neural Network to Tackle Discontinuity in Airway Segmentation

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Manual airway segmentation in CT scans is time-consuming and subjective.
• Automatic segmentation methods struggle with small airway branches and data imbalance.
• Existing methods often yield discontinuous or false-negative predictions for lung diseases.

Purpose of the Study:

• To develop an efficient and robust method for automatic airway segmentation.
• To improve the spatial continuity and accuracy of airway segmentation in CT images.
• To address challenges posed by small airway branches and heterogeneous features.

Main Methods:

• Proposed a novel fuzzy attention neural network (FANN) integrating deep attention and fuzzy logic.
• Introduced a channel-specific fuzzy attention mechanism to handle feature heterogeneity.
• Developed a comprehensive loss function to enhance spatial continuity.
• Formulated a deep fuzzy set with a learnable Gaussian membership function.

Main Results:

• The FANN method demonstrated enhanced spatial continuity and completeness in airway segmentation.
• The approach proved efficient, generalizable, and robust across diverse lung disease datasets (lung cancer, COVID-19, pulmonary fibrosis).
• A novel evaluation metric was proposed and utilized to assess segmentation quality.

Conclusions:

• The proposed FANN method offers an escalated solution for accurate and robust airway segmentation.
• This technique can significantly aid in the examination, diagnosis, and prognosis of various lung diseases.
• The integration of fuzzy logic and attention mechanisms shows promise for complex medical image analysis.