Using the logarithm of odds to define a vector space on probabilistic atlases

Area of Science:

• Medical Imaging
• Computational Anatomy
• Biostatistics

Background:

• Logarithm of the odds ratio (LogOdds) is a probability representation used in various fields.
• Probabilistic atlases are crucial for medical image analysis.
• Existing shape representations may not fully capture anatomical uncertainty.

Purpose of the Study:

• To introduce LogOdds for representing probabilistic atlases in a linear vector space for medical imaging.
• To demonstrate the utility of LogOdds in encoding shape and uncertainty.
• To evaluate the performance of this new representation in image segmentation.

Main Methods:

• Placing probabilistic atlases into a LogOdds linear vector space.
• Relating signed distance maps to LogOdds and comparing with Gaussian smoothing.
• Mapping multiple label maps to capture boundary uncertainty.
• Developing a framework for non-convex interpolations among atlases.
• Generating a deformable shape atlas using principal component analysis in LogOdds space.
• Integrating the atlas into a Bayesian classification segmentation approach for MR images.

Main Results:

• LogOdds representation better preserves shapes in complex, multi-object settings compared to Gaussian smoothing.
• The approach effectively captures uncertainty in boundary locations.
• A deformable atlas generated using LogOdds-based PCA improved segmentation accuracy.
• The Bayesian classification model with LogOdds outperformed standard methods in segmenting subcortical structures on MR images.

Conclusions:

• LogOdds provides a powerful framework for probabilistic atlases in medical imaging, offering superior shape and uncertainty representation.
• This method enhances anatomical modeling and image segmentation, particularly for complex structures.
• The LogOdds space enables natural probabilistic interpretations of vector space operations, facilitating advanced atlas manipulation.