From model based to learned regularization in medical image registration: A comprehensive review.

Area of Science:

• Medical Imaging
• Computer Vision
• Computational Anatomy

Background:

• Image registration is vital for medical applications like disease tracking and treatment planning.
• Accurate deformation capture relies on solving optimization problems, often requiring regularization due to inherent ill-posedness.
• Existing regularization methods for image registration are diverse but lack a unified structure, leading to underutilization.

Purpose of the Study:

• To systematically review and categorize existing regularization methods in medical image registration.
• To introduce a novel taxonomy for understanding and applying regularization techniques.
• To explore the emerging area of learned regularization and its potential.

Main Methods:

• Comprehensive literature review of regularization techniques in medical image registration.
• Development of a novel taxonomy to classify diverse regularization approaches.
• Analysis of the transferability of methods between conventional and deep learning-based registration.

Main Results:

• A structured taxonomy categorizing regularization methods for image registration.
• Identification of learned regularization as a significant emerging trend.
• Examination of challenges and future research directions in the field.

Conclusions:

• Regularization is a critical, yet often overlooked, component of effective image registration.
• A systematic approach and novel taxonomy can guide the selection and development of regularization strategies.
• Further research into learned regularization and method transfer is essential for advancing medical imaging.