Abstract P115: Prediction of Severity of Blood Volume Loss Using ECG Features Based on P, QRS, and T Waves
Introduction: Signal processing and machine learning (ML) paradigms may serve as an automated tool for analyzing and discovering the hidden changes in the ECG in order to predict the presence and severity of hemorrhage. In this study a method based on discrete wavelet transform (DWT) and ML is used to analyze raw ECG, extract significant ECG features, and then use them to identify severe volume loss.
Methods: The algorithm detects and extracts time-interval and amplitude features from the ECG that describe ECG changes due to volume loss. These features include PQ, PR, QT, and ST duration, amplitude of T over amplitude of R, and amplitude of P over amplitude of R from each ECG cycle. The system was tested on volunteers undergoing lower body negative pressure (LBNP) as a hemorrhage mimetic. The LBNP protocol consisted of a 5 min rest period (0 mm Hg) followed by 5 min of chamber decompression of the lower body to −15, −30, −45, and −60 mm Hg and additional increments of −10 mm Hg every 5 min until cardiovascular collapse. Because there is a wide range of LBNP levels resulting in collapse, volume loss was divided into 2 classes: severe (collapse stage together with preceding LBNP stage), and nonsevere as the remaining LBNP stages. This challenges the algorithm to be more individually specific. For each test, only the last 2 min out of each 5 min LBNP stage were used. Extracted features are then transformed with DWT using Daubechies 4 (db4) at level 4. Computed features, without using any baseline ECG for each subject, are used as input to the ML algorithm SVM.
Results: Experimentation occurred in 59 cases with a total of 307 examples. There were 169 (55%) examples for the nonsevere grouping and 138 (45%) for the severe grouping. Overall accuracy of the classification was 89%. The sensitivity and specificity of the severe group were 83.7% and 93.9%, respectively.
Conclusions: Using short interval ECG tracings without baseline ECG comparison, the proposed method was found to accurately detect and classify severe volume loss from nonsevere volume loss in response to LBNP. The method may be suitable to explore designing diagnostics for hemorrhage in which, instead of using only heart rate variability, more detailed ECG information is also extracted and used for identification and severity scoring.