Probabilistic Model for Chronic Obstructive Pulmonary Disease Diagnosis and Phenotyping using Bayesian Network

Amos Otieno Olwendo, Leila Shahmoradi, and Khosrow Agin


Bond Graph Modeling, Mathematical Modeling, Bayesian Network, COPD


World Health Organization (WHO) reports Chronic obstructive pulmonary disease (COPD) as a leading cause of death worldwide with 90\% of COPD deaths realized in middle and low-income countries. COPD prevalence is expected to increase by more than 30\% in the next decade unless urgent actions are taken to reduce the associated risk factors. Reports from Canada, Australia, and United Kingdom show that COPD is prone to misdiagnosis for Asthma yet such countries have hospitals equipped with tools and staff to handle COPD. A spirometer is the main diagnostic tool used to diagnose and separate COPD and Asthma. However, the use of spirometry comes with a number of challenges that make them less useful. This contrast triggers the desire to know the level of COPD prevalence in developing countries in addition to the design a diagnostic device that could be used in resource constrained societies. Consequently, this research seeks to provide a model for COPD diagnosis and classification of cases into phenotypes using a Bayesian Network. Model construction was achieved through developing the Bayesian Network structure and instantiating the parameters for each variable. Model performance is validated using neural network application based on the Levenberg- Marquardt algorithm. Results show that a Bayesian Network has successfully differentiate COPD from Asthma and classified COPD cases into phenotypes. In addition, this study seeks to determine the software requirements for the design and development of a medical device for COPD diagnosis. Such a tool may be used for targeted case finding in resource constrained communities for early identification and treatment of COPD.

Important Links:

Go Back