Use of a Kohonen Neural Network to Characterize Respiratory Patients for Medical Intervention

Abstract

Chronic Obstructive Pulmonary Disease (COPD) is one of theleading causes of respitatory hospilisations in adults inthe USA. Prognosis correlates highly with earlydiagnostics, however the disease may go unnoticed in itsearly stages. A database of 25,000 individuals withrespitatory problems was received for furtherinvestigation. The reported rate of COPD in this populationwas 5.8%, which is fairly low. An unsupervised neuralnetwork using the Kohonen architecture was applied to thedata in order to cluster patients into groups based on riskfactors for COPD. The network consisted of five outputneurons. After training characteristics of the groups wereexamined. Three of the groups consisted of patients withhigh percent of risk factors of COPD. Patients in two ofthose gruoups were correctly diagnosed as having COPD, butpatients in the third group were underdiagnosed for COPD.These patients should be re-examined by a pulmologist forpossible treatment of COPD. Thus Kohonen neural networksmay be a useful tool for clustering patients into groupsfor differential medical intervention.