NR_bloodlosssvm.zip

PPG recordings were acquired from two independent groups of participants: trauma patients with suspected hemorrhage admitted to the hospital, and healthy volunteers subjected to blood withdrawal of 0.9 L. Two minute PPG sequences at various time instances were extracted from each recording, and analyzed using the variable frequency complex demodulation algorithm. As a result, every recording was represented with a novel set of 4 features: goodness of fit, slope of the trend line, the percentage change and the absolute change between estimated amplitude values in the heart rate frequency range at the first and last time points.

In addition, every recording was assigned to one of the classes (blood loss/no blood loss) by 2 or 3 physicians who were blinded to the results of the aforementioned analysis. The automatic classification of blood loss was performed by support vector machines.