Causal Bayesian Networks for Medical Diagnosis: A Case Study in Rheumatoid Arthritis

EasyChair Preprint 4625

Abstract

Bayesian network (BNs) models have been widely applied in medical diagnosis. These models can be built from different sources, including both data and domain knowledge in the form of expertise and literature. Although it might seem simple to depend only on data, this will not be the best approach unless a large dataset is available. In this study, we present a knowledge-based BN modelling approach which we applied for diagnosing the chronic disease of rheumatoid arthritis (RA). We illustrate the process of extracting the relevant knowledge, starting by identifying the BN variables implied by the activities and decision points shown in a model of the care pathway for RA diagnosis. To complete this, further medical knowledge is elicited from an expert panel of rheumatologists, the medical literature is investigated, and a data set is used to parameterize the model. We compare the performance of this knowledge-based BN with another BN model learnt entirely from data. The results show that our proposed knowledge-based model outperforms the data-driven one.

Keyphrases: Bayesian networks, Rheumatoid Arthritis, chronic diseases, diagnosis

@booklet{EasyChair:4625,
  author    = {Ali Fahmi and Amy MacBrayne and Evangelia Kyrimi and Scott Mclachlan and Frances Humby and William Marsh and Costantino Pitzalis},
  title     = {Causal Bayesian Networks for Medical Diagnosis: A Case Study in Rheumatoid Arthritis},
  howpublished = {EasyChair Preprint 4625},
  year      = {EasyChair, 2020}}