IMAGE REGISTRATION WITH OPTIMAL REGULARIZATION PARAMETER SELECTION BY LEARNED AUTO ENCODER FEATURES.

Area of Science:

• Medical Imaging
• Computational Biology
• Image Processing

Background:

• Non-rigid image registration is crucial for analyzing medical images, particularly histopathology.
• Optimizing regularization parameters is essential for accurate and reliable image registration.
• Free Form Deformation (FFD) offers a flexible approach to non-rigid registration.

Purpose of the Study:

• To propose and validate a novel method for optimizing the regularization parameter in FFD non-rigid image registration.
• To enhance the accuracy and regularity of image registration using autoencoder-derived image representations.
• To improve the fine-tuning of histopathology microscope images.

Main Methods:

• Utilized autoencoder-generated image representations to assess generalization quality based on the regularization parameter.
• Incorporated both pixel intensity and learned features to improve the inverse problem solution.
• Implemented the new selection criterion within a multi-level B-spline FFD registration framework with L2-regularization.

Main Results:

• Demonstrated improved accuracy and regularity in the registration process.
• Validated the method's efficacy on both synthetic and real histopathology image datasets.
• Achieved successful fine-tuning of histopathology microscope images.

Conclusions:

• The developed method effectively optimizes the regularization parameter for FFD non-rigid image registration.
• The approach enhances the analysis of histopathology images through improved registration accuracy.
• This technique offers a valuable tool for researchers working with high-resolution microscopy data.