Each year, almost half of all diagnosed primary brain tumors in the United States are Grade IV glioblastoma multiforme (GBMs). While recent efforts have begun to uncover the genetic pathways involved in this cancers etiology and potential methods for treatment arguably, no specific prognostic model has arisen (and been sufficiently validated) to provide widespread usability and individually tailored predictions about a patients prognosis, let alone suggest optimal treatment. The objective of this proposal is the development of a Bayesian belief network (BBN) for predicting outcomes for GBM patients. This research effort first looks to validate a BBN developed using two NIH datasets: the National Cancer Institute (NCI) Rembrandt Project;and the Cancer Genome Atlas (TCGA). Working with our clinical investigators, we will develop a BBN topology for representing GBMs and compute the required conditional probabilities from these public resources. Model variables will encompass the full spectrum of available observations (demographics, initial presentation, histopathology, treatment, imaging, performance scores, end outcomes, etc.).This Rembrandt-TCGA BBN will then be evaluated against conventional statistical models (e.g., multivariate logistic regression, Cox proportional-hazards regression) to assess predictions of time to prognosis and time to survival. Next, we will compare the Rembrandt-TCGA BBN?s performance against two new and unseen populations at UCLA and the Greater Los Angeles Veterans Administration (GLA-VA);this assessment will inform to what extent the nationally-derived BBN can be applied to different GBM populations at other institutions. We also investigate the extension of the Rembrandt-TCGA BBN to add site-specific variables in order to improve prognostic predictions, including symptoms, sequelae, and additional quantitative imaging findings. Informatics-driven tools from our past work in neuro-oncology and image processing will be leveraged to assist with information extraction tasks needed to map UCLA and GLA-VA patient records to site-specific BBNs. Patient- specific models for the UCLA and GLA-VA GBM patients will also be explored using Lazy Bayes Rules to improve modeling and predictive accuracy. The BBNs will also be extended to capture temporal information, resulting in a dynamic belief network (DBN) that represents specific phases of the cancer's evolution and treatment. A graphical user interface (GUI) will permit users to interact with the various belief network models, aiding in medical decision-making tasks. The proposed GUI will allow a clinician to pose questions from either a set of common clinical queries or to create new queries: loading a patient's medical record into this application will automatically populate BBN variables with extracted information, compute the probabilities associated with specified outcomes, and provide a method for similar case-based retrieval. The ultimate result of this endeavor will be a set of tools and well-validated disease model to provide more tailored information and guidance to physicians about GBM patient outcomes.

Public Health Relevance

Grade IV brain tumors, glioblastoma multiforme (GBM), are responsible for a significant number of cancer-related deaths each year. Recent progress in understanding the contributing factors to brain tumor development at the genetic level, its presentation and response to treatment at the imaging level, and improved knowledge of the interplay of chemotherapeutic and other interventions now provide a unique opportunity to create a comprehensive disease model that can aid in medical decision-making tasks. The focus of this research is the creation of a Bayesian belief network for GBMs, using the array of data routinely acquired in the diagnosis and treatment of the problem to provide better predictions about survival outcomes and to develop tools that can help guide physicians with medical decisions regarding treatment options.

National Institute of Health (NIH)
National Cancer Institute (NCI)
Research Project (R01)
Project #
Application #
Study Section
Biomedical Computing and Health Informatics Study Section (BCHI)
Program Officer
Mariotto, Angela B
Project Start
Project End
Budget Start
Budget End
Support Year
Fiscal Year
Total Cost
Indirect Cost
University of California Los Angeles
Schools of Medicine
Los Angeles
United States
Zip Code
Smedley, Nova F; Hsu, William (2018) USING DEEP NEURAL NETWORKS FOR RADIOGENOMIC ANALYSIS. Proc IEEE Int Symp Biomed Imaging 2018:1529-1533
Shen, Shiwen; Han, Simon X; Petousis, Panayiotis et al. (2017) A Bayesian model for estimating multi-state disease progression. Comput Biol Med 81:111-120
Katrib, Amal; Hsu, William; Bui, Alex et al. (2016) ""RADIOTRANSCRIPTOMICS"": A synergy of imaging and transcriptomics in clinical assessment. Quant Biol 4:1-12
Rios Piedra, Edgar A; Taira, Ricky K; El-Saden, Suzie et al. (2016) Assessing Variability in Brain Tumor Segmentation to Improve Volumetric Accuracy and Characterization of Change. IEEE EMBS Int Conf Biomed Health Inform 2016:380-383
Hsu, William; Gonzalez, Nestor R; Chien, Aichi et al. (2015) An integrated, ontology-driven approach to constructing observational databases for research. J Biomed Inform 55:132-42
Singleton, Kyle W; Speier, William; Bui, Alex A T et al. (2014) Motivating the additional use of external validity: examining transportability in a model of glioblastoma multiforme. AMIA Annu Symp Proc 2014:1930-9
Singleton, Kyle W; Bui, Alex A T; Hsu, William (2014) Transfer and transport: incorporating causal methods for improving predictive models. J Am Med Inform Assoc 21:e374-5
Bui, Alex A T; Hsu, William; Arnold, Corey et al. (2013) Imaging-based observational databases for clinical problem solving: the role of informatics. J Am Med Inform Assoc 20:1053-8
Arnold, Corey W; McNamara, Mary; El-Saden, Suzie et al. (2013) Imaging informatics for consumer health: towards a radiology patient portal. J Am Med Inform Assoc 20:1028-36
Singleton, Kyle W; Hsu, William; Bui, Alex A T (2012) Comparing predictive models of glioblastoma multiforme built using multi-institutional and local data sources. AMIA Annu Symp Proc 2012:1385-92