Clustering Based Analysis of Spirometric Data Using Principal Component Analysis and Self Organizing Map

Spirometry is a valuable tool used for respiratory diagnoses and assessment of disease progression. It measures air flow to help make a definitive diagnosis of pulmonary disorder and confirms presence of airway obstruction. In this work, clustering based classification of spirometric pulmonary function data has been attempted using Principal Component Analysis (PCA) and Self Organising Map (SOM). Pulmonary function data (N=100) are obtained from normal and obstructive subjects using gold standard Spirolab II spirometer. These data are subjected to PCA to extract significant parameters relevant to the cluster structure. The clustering analysis of the significant spirometric parameters is further enhanced using self organizing map and classification of spirometric data is achieved. It is observed from results that FEV1, PEF and FEF25 '—' 75% are found to be significant in differentiating normal and obstructive subjects. SOM based classification is able to achieve accuracy of 95%. This cluster based method of feature reduction and classification could be useful in assessing the pulmonary function disorders for spirometric pulmonary function test with large dataset.