Automatic Detection of Dehydration using Support Vector Machines

There is a high demand for techniques that can detect dehydration automatically and accurately. In this study we collected photoplethysmographic (PPG) signals with miniature, wearable pulse oximeters from dehydrated patients being treated in the emergency department of tertiary care medical center. We used a set of features based on the variable frequency complex demodulation (VFCDM) to track changes in the amplitudes of the PPG recordings in the heart rate frequency range over time. These features were fed to support vector machines (SVM) with radial basis function (RBF) kernel for automatic classification. The optimal overall accuracy for classifying dehydration, sensitivity and specificity were 67.91%, 72.77% and 64.31% respectively. These results are promising, and suggest that automatic distinction between dehydration and rehydration is potentially possible even in clinical setting.