An advanced LAN model based on optimized feature algorithm: Towards hypertension interpretability

Abstract

BackgroundThe existing cuffless BP measuring methods still show the inadequacy of the model that utilizes long term dependencies in BP dynamics and come up with the outcome of recurrent calibration. ObjectiveThis study aims to develop a learning-based predictive model for continuous BP measurement with long-term evaluation of the cardiovascular parameter. MethodsTowards hypertension interpretability, the proposed model assists the current as well as the previous status of cardiovascular events based on long short-term memory networks (LSTM) and flexible autoregressive integrated moving average (ARIMA). It is often complicated and challenging to continuously estimate BP from random cardiac events. This paper proposes a new SGFA (Sequential Genetic Feature Algorithm) to tackle the feature optimization problem. ResultsThe experiment was performed on a MIMIC database containing 2780 datafiles of PPG-BP. The proposed model provides the best performance with root mean square error (RMSE) and mean absolute error (MAE) of 1.17 and 1.04 for SBP, whereas 1.06 and 1.02 for DBP. The proposed method has also been evaluated on hypertensive and hypotensive patients. In hypertension condition, estimated SBP and DBP present a good correlation with the true measurements. MAE and RMSE of the estimated SBP are 0.96 and 1.21, whereas, for DBP, it shows 0.42 and 0.57 respectively. Extensive experimentation results confirm that the proposed method delivers a remarkable performance of BP prediction. ConclusionThe proposed work shows a remarkable BP estimation performance compared to the previous inventive methods and signifies insight of various cardiovascular events responsible for BP variation and hypertension interpretability.