Bayesian Hierarchical Models for Voxel-Wise Classification of Prostate Cancer Using Nearest-Neighbor Gaussian Process

• Activity Number: 364 - Contributed Poster Presentations: Section on Medical Devices and Diagnostics
• Type: Contributed
• Date/Time: Tuesday, July 31, 2018 : 10:30 AM to 12:20 PM
• Sponsor: Section on Medical Devices and Diagnostics
• Author(s): Jin Jin* and Joseph Koopmeiners and Gregory Metzger and Ethan Leng
• Companies: Division of Biostatistics, University of Minnesota and Division of Biostatistics, University of Minnesota and University of Minnesota and University of Minnesota
• Keywords: Bayesian hierarchical models; spatial modeling; detection of prostate cancer; Nearest Neighbor Gaussian Process; multi-parametric MRI

Abstract

Multi-parametric MRI (mpMRI) is increasingly popular as a non-invasive method for detecting prostate cancer (PCa). Current classification algorithms use mpMRI biomarkers as predictors and provide PCa detection of similar accuracy. However, these existing methods cannot systematically account for other informative features in the data that potentially improve PCa detection. One important feature is strong spatial correlation. However, classical spatial models are computationally infeasible due to the size of the data. This project aims to develop a novel spatial model to improve voxel-wise classification of PCa by accounting for spatial correlation and patient-specific effects in the mpMRI data. We use Nearest Neighbor Gaussian Process (NNGP) as the prior for spatial random effects, which ensures computational efficiency by approximating the full spatial model with an alternative that only considers spatial correlation among the neighboring voxels. Simulation studies and real data analyses show that our proposed model achieves high classification accuracy with significant improvement compared to current non-spatial models and commonly used Gaussian Markov Random Field models.