Fast three-dimensional image generation for healthy brain aging using diffeomorphic registration.

Area of Science:

• Neuroimaging
• Medical Image Analysis
• Computational Neuroscience

Background:

• Longitudinal magnetic resonance imaging (MRI) cohorts are crucial for understanding brain aging and neurodegenerative diseases.
• Missing data in these cohorts hinders comprehensive analysis and accurate prognosis.
• Existing methods may not preserve anatomical details crucial for clinical applications.

Purpose of the Study:

• To develop a methodology for generating anatomically plausible MRI scans to fill missing data in longitudinal brain aging studies.
• To simulate subject-specific aging processes using deep learning-based diffeomorphic registration.
• To enhance the utility of longitudinal neuroimaging datasets for research and clinical applications.

Main Methods:

• Utilized Synthmorph, a deep learning-based diffeomorphic registration method, with novel modules to simulate 3D brain aging between initial and final MRI scans.
• Employed six image similarity metrics to align generated images to specific age ranges.
• Assumed linear brain decay in healthy subjects to estimate the age of synthetic images.

Main Results:

• Generated 7548 synthetic 3D MRI scans across three longitudinal cohorts (ADNI, OASIS-3, GENIC), simulating scans every six months.
• Achieved state-of-the-art results in quantitative and qualitative assessments, validated by a neuroradiologist.
• Confirmed the accuracy of the linear brain decay assumption (R² ∈ [.924, .940]).

Conclusions:

• The proposed methodology effectively generates anatomically plausible brain aging predictions, enhancing longitudinal datasets.
• Diffeomorphic registration preserves brain anatomy better than other generative methods, making it suitable for clinical use.
• This approach efficiently simulates 3D MRI scans of brain aging from two time points.