Generative diffeomorphic modelling of large MRI data sets for probabilistic template construction.

Area of Science:

• Medical Imaging
• Computational Anatomy
• Bayesian Modeling

Background:

• Medical image analysis requires robust methods for handling population variability in anatomical shapes and signal intensities.
• Automated learning of probabilistic tissue templates is crucial for standardizing medical imaging studies.
• Existing medical image registration tools face challenges in accurately aligning complex structures across diverse populations.

Purpose of the Study:

• To introduce a novel hierarchical generative model for medical image data.
• To demonstrate the model's capability in capturing anatomical and signal intensity variability.
• To apply the model for automated learning of probabilistic tissue templates and validate its performance in neuroimaging.

Main Methods:

• Development of a hierarchical generative Bayesian model for medical image data.
• Application of the model to real and synthetic brain Magnetic Resonance (MR) scans, including the cervical cord.
• Validation against state-of-the-art medical image registration techniques.

Main Results:

• The model accurately captures signal intensity and anatomical shape variability across large populations.
• It achieves accurate alignment of brain and spinal cord structures, comparable to current leading methods.
• Generated tissue probability maps facilitate automated segmentation, bias correction, and spatial normalization of unseen MR data.

Conclusions:

• The proposed hierarchical generative model offers a powerful, automated solution for learning probabilistic tissue templates.
• It provides accurate registration and essential pre-processing capabilities for neuroimaging studies.
• The Bayesian approach is broadly applicable to various medical image computing challenges.