Classification of normal and abnormal respiration patterns using flow volume curve and neural network

Abstract

Lung diseases affect many people's lives. Early and correct diagnosis of respiratory system abnormalities is vital to patients. While spirometry is the most common pulmonary function test, the interpretation of the results is dependent on the physicians' experience. A decision support system can help physicians in correct diagnoses. This study aims at designing a system for detecting pulmonary system normal and abnormal functions by using spirometry data and multilayer perceptron neural networks (MLPNN). To detect and classify respiratory patterns into normal, obstructive, restrictive and mixed patterns, curves are fitted to flow-volume data of the patients. The fitted curve coefficients and predicted values for FEV1, FVC, and FEV1% are used as inputs to the MLPNN. Different MLP structures were tested. The spirometric data were obtained from 205 adult volunteers. Total accuracy, sensitivity and specificity among the four categories are 97.6%, 97.5% and 98.8% respectively.