Simultaneous Longitudinal Registration with Group-Wise Similarity Prior

Area of Science:

• Medical Image Analysis
• Computational Anatomy
• Neuroimaging

Background:

• Longitudinal neuroimaging studies are crucial for understanding disease progression.
• Accurate registration of serial brain scans is essential for quantifying anatomical changes.
• Existing methods may not fully leverage shared information across multiple patient scans.

Purpose of the Study:

• To develop a novel group-wise registration algorithm for simultaneous processing of multiple longitudinal image pairs.
• To enhance the accuracy and biological relevance of anatomical change quantification in neurodegenerative diseases.
• To improve the prediction of future anatomical states using learned transformation models.

Main Methods:

• Implementation of a group-wise consistency prior within the Large Deformation Diffeomorphic Metric Mappings (LDDMM) framework.
• Simultaneous registration of N longitudinal image pairs, constraining individual transformations to be similar to an average.
• Utilizing geodesic shooting to model large deformations and maintain group-wise consistency.

Main Results:

• The group-wise prior strengthens the common anatomical signal across N image pairs.
• Momenta learned with the group-wise prior demonstrated improved prediction of a third, unobserved time point.
• The algorithm was successfully tested on 57 longitudinal Alzheimer's Disease patient image pairs from the ADNI database.

Conclusions:

• Group-wise registration effectively couples longitudinal image pairs, improving registration accuracy.
• This method enhances the representation of long-term biological processes in anatomical changes.
• The algorithm shows potential for more accurate disease progression modeling and outcome prediction in Alzheimer's Disease.