Abstract 2: Prediction of Hypovolemia Severity Using ECG Signal with Wavelet Transformation Analysis From a Mobile Armband
Introduction: While ECG analysis has the potential to assist in tracking volume loss, issues regarding location of measured signal as well as whether accuracy is degraded without normalization to baseline remain challenges for point-of-care utilization. We used discrete wavelet transform (DWT) based methods to analyze ECG signal measured by a mobile electronic armband, to detect the presence of the severe hypovolemia.
Methods: ECG was captured with a wireless electronically enabled elastic armband attached to the left arm of 22 volunteers undergoing lower body negative pressure (LBNP). The LBNP protocol consisted of a 5-minute rest period (0 mm Hg) followed by 5 minutes of chamber decompression of the lower body to -15, -30, -45, and -60 mm Hg and additional increments of -10 mm Hg every 5 minutes until cardiovascular collapse. Armband ECG was sampled at 128 Hz. LBNP levels were divided into 2 classes: severe (collapse stage together with preceding LBNP stage), and non-severe as the remaining LBNP stages. DWT features are introduced to develop a classification scheme to determine the severity of volume loss. Volume severity predictions are compared (with and without baseline normalization). The machine-learning algorithm SVM with leave-one-out cross-validation was used for classification.
Results: 136 classification samples were used. For classification with features normalized to baseline values, the prediction accuracy was 86%. The sensitivity and specificity for prediction of the severe class were 80.6% and 90.5%, respectively. For classification without normalization to baseline, SVM provides 81.6% accuracy. The sensitivity and specificity for prediction of the severe class were 75.8% and 86.5%, respectively.
Conclusion: DWT may have the ability to rapidly determine the severity of acute volume loss. This study also shows that DWT of the ECG may not require baseline values and normalization to classify volume loss with reasonable accuracy, sensitivity, and specificity making it useful for point-of-care use. The feasibility of processing the ECG collected by a mobile armband at low frequency is also demonstrated. This may be helpful for remote wireless monitoring of many types of subjects including warfighters, civilians, and mass casualties.