Groupwise image registration based on a total correlation dissimilarity measure for quantitative MRI and dynamic...

Area of Science:

• Medical Imaging
• Image Registration
• Computational Anatomy

Background:

• Pairwise image registration, a common technique, requires individual optimization for each image pair.
• Mutual information is a widely used dissimilarity measure in pairwise registration.
• Groupwise registration methods aim to register multiple images simultaneously without a fixed reference.

Purpose of the Study:

• To adapt a multivariate mutual information measure, total correlation, for groupwise medical image registration.
• To provide a theoretically grounded approach for groupwise registration.
• To compare the performance of total correlation with existing groupwise methods.

Main Methods:

• Adapted total correlation, a multivariate mutual information measure, for groupwise image registration.
• Implemented and detailed the total correlation measure for this application.
• Conducted experiments on five quantitative imaging datasets and one dynamic CT dataset.

Main Results:

• Total correlation yielded registration results comparable to Huizinga's principal component analysis-based methods.
• The proposed total correlation measure demonstrated strong performance across various datasets.
• Results showed total correlation is a viable alternative to pairwise mutual information for quantitative imaging.

Conclusions:

• Total correlation provides a theoretically justified and effective method for groupwise medical image registration.
• This approach offers comparable performance to empirical methods like Huizinga's.
• Total correlation presents a valuable alternative for registering quantitative imaging datasets.