MRF-based deformable registration and ventilation estimation of lung CT

Area of Science:

• Medical Image Analysis
• Computational Imaging
• Radiology

Background:

• Deformable image registration is crucial for medical image analysis, particularly for lung CT.
• Lung CT registration faces challenges like large motion, sliding motions, and contrast changes.
• Existing discrete optimization methods (MRF) offer advantages but often reduce accuracy due to simplifications.

Purpose of the Study:

• To develop a novel discrete optimization strategy for lung CT registration that overcomes limitations of previous methods.
• To enhance registration accuracy and computational efficiency.
• To enable direct estimation of lung ventilation.

Main Methods:

• Utilized an image-derived minimum spanning tree for efficient graph-based optimization, adept at handling sliding motions.
• Introduced a stochastic sampling approach for image similarity within a diffeomorphic B-spline transformation model with diffusion regularization.
• Incorporated hyper-labels alongside geometric transform labels to represent local intensity variations for ventilation estimation.

Main Results:

• The novel approach significantly improves registration accuracy and performance on exhale-inhale CT volume pairs.
• The minimum spanning tree structure efficiently optimizes complex sliding motions.
• The method achieves reduced computational complexity, enabling larger label space minimization and accurate lung ventilation estimation.

Conclusions:

• The proposed discrete optimization method enhances lung CT registration accuracy and efficiency.
• The technique effectively addresses key challenges in lung CT registration, including motion and contrast variations.
• This work provides a foundation for improved medical image analysis and direct lung ventilation estimation.