Registration of longitudinal brain image sequences with implicit template and spatial-temporal heuristics

Area of Science:

• Neuroimaging
• Medical Image Analysis
• Computational Anatomy

Background:

• Accurate measurement of longitudinal changes in brain structures and functions is crucial for clinical studies.
• Comparing longitudinal changes across subjects is vital for identifying disease-related alterations.
• Current methods face challenges in simultaneously addressing within-subject and across-subject longitudinal analysis.

Purpose of the Study:

• To develop a novel method for simultaneously registering longitudinal image sequences from multiple subjects to a common space.
• To enable consistent measurement of longitudinal changes within each subject.
• To facilitate joint alignment of all subjects' image sequences to a hidden common space without an explicit template.

Main Methods:

• Introduction of temporal fiber bundles to model spatial-temporal behavior of anatomical changes.
• Development of a probabilistic model based on temporal fibers for spatial smoothness and temporal continuity.
• Simultaneous estimation of transformation fields using the expectation maximization (EM) algorithm and maximum a posteriori (MAP) estimation.

Main Results:

• The proposed method successfully measures longitudinal brain changes, demonstrated by hippocampus volume changes.
• Achieved superior performance compared to traditional pairwise or groupwise registration methods.
• Enabled consistent within-subject change measurement and joint across-subject alignment.

Conclusions:

• The novel simultaneous registration method effectively addresses the challenge of measuring and comparing longitudinal brain changes.
• This approach enhances the identification of disease-related changes by improving the accuracy of neuroimaging analysis.
• The method shows significant potential for application in various clinical studies involving longitudinal neuroimaging data.