Generative Method to Discover Genetically Driven Image Biomarkers.

Area of Science:

• Computational biology
• Medical imaging analysis
• Genetics and genomics

Background:

• Many diseases present with multiple simultaneous pathologies, complicating accurate quantification.
• Integrating diverse data types like imaging and genetics is crucial for a comprehensive understanding of disease.
• Existing methods may not fully capture the complex interplay between genetic factors and disease presentation.

Purpose of the Study:

• To develop a generative probabilistic model for discovering disease subtypes based on genetic variants.
• To jointly model and analyze co-occurring patterns in medical images and genetic data.
• To quantify disease heterogeneity in patients by identifying underlying subtypes.

Main Methods:

• Utilized a generative probabilistic approach based on a variant of topic models.
• Developed an efficient variational inference algorithm for pattern extraction.
• Jointly modeled image and genetic markers to identify latent disease structures.

Main Results:

• The method successfully identifies common co-occurring image and genetic patterns.
• Quantified the presence of heterogeneous disease processes within individual patients.
• Demonstrated utility in characterizing subtypes of Chronic Obstructive Pulmonary Disease (COPD).

Conclusions:

• The proposed generative model effectively discovers disease subtypes by integrating imaging and genetic data.
• This approach provides a robust framework for quantifying disease heterogeneity.
• The findings have significant implications for personalized medicine and disease subtyping, as illustrated in COPD research.