Biomedical Signal Processing and Control · 2019 · 36 citations · 44 references
ComputingCardiologyHealth InformaticsBiomedical Signal ProcessingClinical InformaticsMachine LearningBlood Pressure Signal AnalysisCardiovascular AssessmentCardiovascular MonitoringSupport Vector MachinesVital Signs AssessmentVital Signs RecordingBlood PressureBlood Pressure EstimationBlood Pressure MonitoringBoostingContinuous Blood Pressure MonitoringDiastolic Blood Pressure EstimationFeature Extraction From Blood Pressure SignalsSystolic Blood Pressure Estimation
Traditionally, blood pressure (BP) is measured by cuff-based instruments, which is inconvenient and does not allow continuous measurement. In fact, continuous blood pressure monitoring is precious for gaining information about the health conditions of people. In this paper, an algorithm is developed using a modified group method of data handling technique to estimate systolic BP (SBP) and diastolic BP (DBP) by only photo-plethysmogram (PPG) signal in a continuous manner. The estimation does not require to calibration and is done without any knowledge about significant features, and features are selected based on their competency. The Multi-parameter Intelligent Monitoring in Intensive Care dataset is used for training the system, and the hold-out validation is used in 7 runs by assuming 70% of the samples as a train set. Moreover, a designed hardware device is used for recording data on 25 subjects for testing the trained model on the real-world. As the state-of-the-art, four popular regression algorithms, including support vector machine, adaptive boosting, decision tree and random forest, are trained with conventional features, which use an electrocardiogram (ECG) and PPG signals simultaneously and compared with the proposed system. The root means square error (RMSE) and the mean absolute error (MAE) of the proposed algorithm is 3.47 ± 1.31 and 2.40 ± 1.01 mmHg for SBP and 4.67 ± 1.44 and 3.33 ± 1.61 mmHg for DBP respectively. Also, the RMSE of the proposed algorithm on recorded data by hardware device is 3.2 ± 0.7 and 4.4 ± 1.0 mmHg, and the MAE is 2.2 ± 0.7 and 2.9 ± 1.2 mmHg for SBP and DBP, respectively. The proposed algorithm is not dependent on synchronization of ECG and PPG, and it is promising compared to the state-of-the-art, especially in recorded data by new hardware.
44
Classification and Regression by randomForest
Andy Liaw, Matthew C. Wiener · 2007 · 18.4K citations
A new optimizer using particle swarm theory
R.C. Eberhart, James Kennedy · 2002 · 14.9K citations
Ridge Regression: Biased Estimation for Nonorthogonal Problems
Arthur E. Hoerl, Robert W. Kennard · Technometrics · 1970 · 8.9K citations
Ridge Regression: Biased Estimation for Nonorthogonal Problems
Arthur E. Hoerl, Robert W. Kennard · Technometrics · 2000 · 6.6K citations