Hierarchical discriminative framework for detecting tubular structures in 3D images

Area of Science:

• Medical Imaging
• Computer Vision
• Radiology

Background:

• Detecting tubular structures (e.g., airways, vessels) is crucial for medical diagnosis and surgical planning.
• Current methods often rely on root-to-tip progression and image filtering, requiring application-specific tailoring and struggling with noise or low contrast.

Purpose of the Study:

• To develop a robust and adaptable algorithm for detecting tubular structures in medical images.
• To overcome limitations of existing methods in noisy and low-contrast environments.

Main Methods:

• A two-layer model combining a low-level likelihood measure and a high-level branch verification.
• Utilizing a discriminative classifier for robust tubular presence detection across multiple scales.
• Employing a multi-scale shortest path algorithm for candidate centerline and radius extraction.
• Implementing a learning-based indicator function to discard false positive branches.

Main Results:

• The proposed technique demonstrates robustness against noise and imaging artifacts.
• Successful detection of airways in rotational X-ray volumes was achieved, even in challenging conditions.
• The algorithm effectively identifies tubular structures without extensive application-specific tuning.

Conclusions:

• The novel two-layer model provides a robust solution for tubular structure detection in medical imaging.
• This approach enhances the reliability of airway detection in rotational X-ray imaging, aiding diagnosis and surgical planning.