Bayesian Modeling of Cancer Outcomes Using Genetic Variables Assisted by Pathological Imaging Data

Area of Science:

• Oncology
• Bioinformatics
• Medical Imaging

Background:

• Genetic profiling is crucial for modeling cancer outcomes, but often lacks sufficient information for accurate predictions.
• Pathological images offer a cost-effective and widely available data source in cancer research and clinical practice.

Purpose of the Study:

• To develop a Bayesian approach for selecting genetic variables and modeling cancer outcomes by integrating pathological imaging features.
• To enhance the performance of predictive models by "borrowing" information from imaging data.

Main Methods:

• A Bayesian framework was employed for variable selection and outcome modeling.
• Information was "borrowed" from low-dimensional pathological imaging features by reinforcing selection results between imaging and outcome data.
• A weighting strategy was developed to manage varying effectiveness of information borrowing across subjects.

Main Results:

• Simulations demonstrated the proposed approach's competitive performance compared to existing methods.
• Analysis of The Cancer Genome Atlas (TCGA) lung adenocarcinoma (LUAD) data revealed novel findings regarding overall survival and gene expression.
• The method produced results distinct from alternatives and exhibited sound statistical properties.

Conclusions:

• Integrating pathological imaging features into genetic analysis offers a promising strategy to improve cancer outcome modeling.
• The proposed Bayesian approach provides a robust and effective method for leveraging multimodal data in cancer research.
• This work highlights the potential of combining imaging and genetic data for more accurate cancer prognostication.