Model Selection for Diagnosis and Treatment Using Temporal Influence Diagrams

This paper describes the model-selection process for a temporal influence diagram formalization of the diagnosis and treatment of acute abdominal pain. The temporal influence diagram explicitly models the temporal aspects of the diagnosis/treatment process as the findings (signs and symptoms) change over time. Given the complexity of this temporal modeling process, there is uncertainty in selecting (1) the model for the initial time interval (from which the process evolves over time); and (2) the stochastic process describing the temporal evolution of the system. Uncertainty in selecting the initial Bayesian network model is addressed by sensitivity analysis. The choice of most appropriate stochastic process to model the temporal evolution of the system is also discussed briefly.