Diffeomorphic Lung Registration Using Deep CNNs and Reinforced Learning.

Area of Science:

• Medical Imaging
• Computer Vision
• Machine Learning

Background:

• Image registration is crucial for analyzing medical scans.
• Comparing chest inspiratory and expiratory CT scans requires precise alignment.
• Existing methods may face challenges with large datasets and accuracy.

Purpose of the Study:

• To develop an accurate and efficient method for registering chest CT scans.
• To recover the diffeomorphic elastic displacement vector field (DVF).
• To improve upon existing image registration techniques using reinforced learning.

Main Methods:

• Utilizing a RegNet-based architecture.
• Implementing a reinforced learning strategy for large datasets.
• Jointly regressing direct and inverse transformations to recover the DVF.

Main Results:

• The proposed method achieved lower estimation error than the standard RegNet approach.
• The reinforced learning strategy effectively handled large training datasets.
• Accurate recovery of the displacement vector field was demonstrated.

Conclusions:

• The reinforced learning approach enhances accuracy in medical image registration.
• This method offers a promising solution for aligning serial medical images.
• Further applications in respiratory motion analysis are suggested.