Cardiovascular disease (CVD) is a common disease nowadays, and hypertension is one of the predictors of it. For unobtrusive BP measurement, continuous blood pressure (BP) estimation using Pulse Transit Time (PTT) is a useful technique. The accuracy is a fact as it is feasible for a large number of phases in our life. So a Deep Neural Network (DNN) based blood pressure estimation, using data mining is proposed in this paper. Instruments were made by ourselves for collecting real data using an electrocardiogram (ECG) and photoplethysmogram (PPG) sensor as the real machine was so much expensive. Fourteen features were extracted, and best features were selected using a genetic algorithm. After this, DNN, Multivariate linear regression (MLR) and Support vector regression (SVR) models were constructed using these features. The accuracy and robustness were validated according to our equipment as the Pearson’s R between referenced and estimated BP for the static experiment (SBP) were 0.90 and for dynamic (DBP) is 0.82 for DNN model. DNN model is compared with MLR and SVR. DNN gave comparatively good results than MLR and SVR and also better than the PTT-based model.
Paper
Full text
Deep Neural Network Based Continuous Blood Pressure Estimation with Data Mining Techniques
Semantic Scholar · Medicine · 2019
Abstract
Cardiovascular disease (CVD) is a common disease nowadays, and hypertension is one of the predictors of it. For unobtrusive BP measurement, continuous blood pressure (BP) estimation using Pulse Transit Time (PTT) is a useful technique. The accuracy is a fact as it is feasible for a large number of phases in our life. So a Deep Neural Network (DNN) based blood pressure estimation, using data mining is proposed in this paper. Instruments were made by ourselves for collecting real data using an electrocardiogram (ECG) and photoplethysmogram (PPG) sensor as the real machine was so much expensive. Fourteen features were extracted, and best features were selected using a genetic algorithm. After this, DNN, Multivariate linear regression (MLR) and Support vector regression (SVR) models were constructed using these features. The accuracy and robustness were validated according to our equipment as the Pearson’s R between referenced and estimated BP for the static experiment (SBP) were 0.90 and for dynamic (DBP) is 0.82 for DNN model. DNN model is compared with MLR and SVR. DNN gave comparatively good results than MLR and SVR and also better than the PTT-based model.