Abstract
Mean Arterial Pressure (MAP) is defined as central pressure in the arteries of a person during a single cardiac cycle. It is regarded as an important bio-marker of blood perfusion in vital organs as compared to systolic blood pressure (SBP). The actual MAP can be determined by manual monitoring and complex calculations limited to occasional monitoring status. Growing personalized health care monitoring devices have already evinced a variety of health parameters to track on a daily basis with the additional advantage of continuous, noninvasive, and unobstructed measurement. This work proposes a direct strategy for the estimation of mean arterial pressure without using the systolic and diastolic BP values. By exploring 13 significant morphological features from a single PPG signal which are most related to the target MAP are derived such as Pulse Interval, Inflection Ratio etc. The estimation is performed using LSTM network with an architecture having 2- LSTM layers followed by a dropout and dense layer. With 942 subjects of UCI repository dataset our model achieves a remarkable mean absolute error of 1.48, standard deviation of 2.36 and pearson correlation coefficient of 0.96 which is better as compared to the existing works and even chalked up the British Hypertension Society (BHS) benchmark with grade A.