Deep Long Short-Term Memory Networks for Asthma Disease Prediction from Respiratory Dataset Obtained from the PEEP Expiratory Occlusion

Abstract

This study presents a system using Long-shortterm memory (LSTM) network for asthma disease prediction from respiratory dataset. For neural network training, a total of 34,000 respiratory data sample collected as a time series from a total of 68 people, containing 200 time step windows, were used. The data was then split into training and test sets, and of the 10,200 test data, 3,273 belong to asthma patients and 6,927 belong to healthy individuals. The LSTM network was provided with input consisting of measurement values of pressure, flow, tidal volume, chest circumference, abdominal circumference, inspiratory indices, global aeration, and aeration data inspiratory indices collected from various measurement devices for the purpose of classifying asthma. This respiratory data set was used for the first time in this study as a deep learning application. Consequently, 88.8% of patients with asthma and 94.4% of persons without any health issues were accurately diagnosed. The overall classification accuracy of the model was obtained as 93%.