Classification of Spirometry Using Stacked Autoencoder based Neural Network

Abstract

Spirometry is the most common and effective way to diagnose various severe respiratory diseases such as Chronic Obstructive Pulmonary Disease (COPD), asthma, occupational lung diseases and pulmonary hypertension. A variety of measurements can be taken and expounded from spirometry; Forced Vital Capacity (FVC), the maximal amount of air one can forcefully exhale in one second (FEV1) and the ratio FEV1/FVC are significant measurements to diagnose the problems with lung functionality (Fig. 1). The objective of this study was to accurately classify the abnormal spirometry using stacked autoencoder (SAE) based neural network by extracting the features from the flow-volume curve. Abnormal spirometry is decided based on the values of FEV1, FVC and the ratio of FEV1/FVC are less than the Lower Limit of Normal (LLN) [1], predicted from the standard reference equations [2].The proposed method shows accuracy of 96.57% for FEV1, 96.01% for FVC and 98.98% for the ratio FEV1/FVC.