Web proceedings papers


Barbora Andrášikova , Ana Madevska-Bogdanova , Milan Tysler , Magdalena Kostoska , Ivan Kuzmanov , Oto Masar , Silvia Putekova , Vladimir Trajkovik and Fedor Lehocki


Blood pressure is one of the most important vital signs which is used to assess patient’s medical state. However, traditional methods for its measurement are not suitable for critical situations such as disasters where it is required to prioritize victims based on their medical state. Given the vital importance of blood pressure, it is necessary to estimate it as accurately as possible without manual calibration. In this work, we evaluated blood pressure estimation methods based on ECG and PPG signals using machine and deep learning algorithms. Best results (mean absolute error of 2.75 mmHg) were achieved with 1 second long ECG segments using a neural network composed of two hidden LSTM layers with the first one being bidirectional. These results were obtained without manual calibration which is not feasible under the aforementioned circumstances.


blood pressure, ECG, PPG, deep learning, machine learning, emergency medicine