A BAYESIAN MARK INTERACTION MODEL FOR ANALYSIS OF TUMOR PATHOLOGY IMAGES

Area of Science:

• Computational pathology
• Statistical modeling
• Cancer biology

Background:

• Digital pathology generates massive high-resolution imaging data for cancer diagnosis.
• Deep learning advances cell identification and classification in pathology images.
• Statistical modeling of spatial cell patterns offers insights into tumor progression and cancer mechanisms.

Purpose of the Study:

• To model spatial correlations among commonly observed cell types in tumor pathology images.
• To develop a novel geostatistical marking model within a Bayesian framework.
• To enhance understanding of cell-cell interactions in cancer progression.

Main Methods:

• Proposed a novel geostatistical marking model with interpretable parameters in a Bayesian framework.
• Employed auxiliary variable Markov chain Monte Carlo (MCMC) algorithms for posterior distribution sampling.
• Applied the model to benchmark datasets and a case study of 188 lung cancer patients.

Main Results:

• The model provides sharper inferences compared to traditional exploratory analyses.
• Spatial correlation analysis between tumor and stromal cells in lung cancer predicts patient prognosis.
• Demonstrated a new method for characterizing spatial correlations in multitype spatial point patterns.

Conclusions:

• The developed statistical methodology offers a novel approach to multitype spatial point pattern analysis.
• Spatial cell-cell interactions, specifically between tumor and stromal cells, are significant predictors of patient prognosis.
• This work provides a new perspective on understanding the role of cell-cell interactions in cancer progression.