Quantitative Assessment of Respiratory Distress Using Convolutional Neural Network for Multivariate Time Series Segmentation

Respiratory Distress (RD) is typically associated with many critical conditions, particularly in aged population. The present-day monitoring solutions do not track the severity of RD distinctly and this obscures accurate prognosis. Our method quantifies RD condition to a ‘Severity Index (SI)’ in a real time monitoring setup. We achieve this by tracking pattern of respiratory rate (RR), oxygen saturation (SpO
2
) and heart rate (HR) over time using Convolutional Neural Network (CNN) based model for multivariate time series segmentation. The CNN based approach is then compared with other two different approaches viz. Long-Short Term Memory (LSTM) model and Bayesian Inference model based on Symbolic Aggregate approXimation (SAX). We work on segments extracted from 180 records of MIMIC-III Clinical and Waveform Database, after getting them clinically annotated. Our CNN, LSTM and SAX models achieve RMSE (Root Mean Square Error) of 0.288, 0.301 and 0.305 respectively with respect to the clinical annotations. All methods achieve AUC (Area Under Curve for Receiver Operating Char-acteristic(ROC)) close to 0.95. The dataset, code and trained models are available at https://github.com/rohit-pardasani/RDSegmentation_v1.