Analysis of restrictive pulmonary function abnormality using spirometric investigations and QPSO feature selection

Restrictive pulmonary disorders are leading cause of morbidity and mortality worldwide. Spirometric pulmonary function test is the most common method used to assess restrictive disorder. The sub-maximal effort of the patients may often lead to misclassification due to large interdependency among the spirometric parameters. Also, there is requirement that too many parameters are to be analysed by the physician. In this study, Quantum-behaved Particle Swarm Optimisation (QPSO) approach has been used for identification of significant features that are useful in disease diagnosis. Then the selected feature set is evaluated based on the error in prediction of FEV1, PEF and FEV6 using radial basis function neural network. Results show that QPSO is able to identify most significant features in both normal and restrictive. It is also observed that logistic model tree classifier achieves accuracy of 95%. This feature selection appears to aid in detection of pulmonary function disorders using spirometric investigations.