Course Level: Beginner
Format Requirements: This is both a conceptual and a hands-on course. Although a beginner-level course, participants should be comfortable with basic notions of probability. Additional theory such as conditional probability and Bayes’ Theorem will be introduced during the course. Participants should bring a Windows/Macintosh-based PC for computational tutorials. A guest license for NETICA and the tutorial examples will be provided to course participants.
Background: Bayesian Network (BN) is a probabilistic graphical model used for data classification, identification of causal relationships, and output prediction. BNs allow a domain expert to model uncertain relationships between a variable of interest with unknown values (e.g., modeling uncertain relationships to predict risk of a disease) and clinical findings/observations (known variables) and are particularly useful for medical diagnosis (e.g., estimating breast cancer risk using mammography findings). Attractive features of BNs include encoding dependencies among all variables, thereby addressing problems with incomplete data; informing causal relationships, thereby increasing understanding about a problem domain and predicting consequences of treatment; combining prior knowledge (which often comes in causal form) and available data; and user friendliness of graphical representations.
Description and Objectives: The course will begin with a general description of BN theory and statistics, demonstrate how BNs differ from decision trees, causal diagrams, and other statistical and data mining techniques such as logistic regression and artificial neural networks. Use of NETICA software for BN construction and its application to various medical decision-making problems will be demonstrated, focusing on breast cancer risk prediction using mammography observations and patient demographic factors. The session will conclude with discussion of limitations and extensions of BNs.
• Understand the theory underlying Bayesian Networks (BNs).
• Understand similarities and differences between BNs and other modeling and analytic techniques.
• Application of BNs to medical decision-making problems.
• Understand BNs’ limitations and potential extensions.