Patient-specific Conditional Joint Models of Shape, Image Features and Clinical Indicators

Area of Science:

• Medical Image Analysis
• Statistical Shape Modeling
• Computational Anatomy

Background:

• Statistical shape modeling and medical image analysis often require integrating diverse data types.
• Existing methods may lack flexibility in handling mixed variable types and complex dependencies.

Purpose of the Study:

• To develop a joint statistical model integrating anatomical shapes, image features, and clinical indicators.
• To enhance statistical shape modeling and medical image analysis using a flexible modeling approach.

Main Methods:

• Employed a copula model to decouple joint dependency structures from marginal distributions.
• Incorporated binary, discrete, ordinal, and continuous variables.
• Utilized Bayesian conditional models with Gaussian processes for dependency structures.

Main Results:

• Demonstrated a flexible method for including various variable types in joint modeling.
• Successfully applied the method to a stroke dataset, modeling lateral ventricles, white matter hyperintensities, and clinical indicators.
• Generated interpretable joint models and patient-specific statistical shape models.

Conclusions:

• The proposed copula-based joint model offers a flexible and powerful approach for integrated analysis of shape, image, and clinical data.
• This method advances statistical shape modeling and medical image analysis, particularly in the context of neurological diseases like stroke.