Multi-feature statistical nonrigid registration using high-dimensional generalized information measures

Area of Science:

• Medical Imaging
• Computational Anatomy
• Radiology

Background:

• Nonrigid image registration is crucial for aligning medical images with complex variations.
• Information-theoretic measures, particularly mutual information, are widely used but have limitations.
• Generalized information measures offer alternatives for specific clinical needs, often focusing on greylevel distributions.

Purpose of the Study:

• To develop and optimize generalized information measures for high-dimensional, multi-feature nonrigid image registration.
• To introduce novel estimators for alpha-informations and study their variational optimization.
• To provide a computationally efficient and well-posed framework as an alternative to entropic graph techniques.

Main Methods:

• Development of consistent and asymptotically unbiased kappa nearest neighbors estimators for alpha-informations.
• Variational optimization of these estimators over smooth transform spaces (finite and infinite dimensional).
• Application and assessment on tagged MRI for myocardial deformation and perfusion MRI for cardio-thoracic motion compensation.

Main Results:

• Novel estimators for alpha-informations were introduced and their optimization studied.
• A theoretically sound and computationally efficient framework for nonrigid registration was established.
• The framework demonstrated effective performance in complex cardiological imaging applications.

Conclusions:

• The proposed framework offers a robust and efficient approach for nonrigid image registration using generalized information measures.
• The novel estimators and optimization techniques advance the field of statistical image analysis in radiology.
• Successful application in cardiac MRI highlights the clinical relevance for deformation and motion analysis.