Metastasis identification based on clinical parameters using Bayesian network
Arida Ferti Syafiandini, Ito Wasito · 2016
Bayesian Network is a promising method for modelling probabilistic relationships among causally related variables. This paper presents an application of Bayesian Network for identifying the occurrence of metastasis in a patient with positive or negative breast cancer tumor based on observed clinical parameters. Its structure is built using K2 search algorithm with a topological order obtained from Minimum Weight Spanning Tree. Maximum Likelihood Estimation is also employed to learn parameter (prior and conditional probabilities) from network structure. For metastasis identification, the marginal probability is computed using Junction Inference Tree. Compared to Logistic Regression and Linear Discriminant Analysis, Bayesian Network gives a clearer idea about how each clinical parameter relates to another. In terms of average accuracy, sensitivity, and selectivity, Bayesian Network also outperforms those methods.