As one of the important physiological parameters of human being, blood pressure (BP) can reflect the status of human cardiovascular function. It is also an important indicator of disease prevention and diagnosis clinically. In recent years, non-invasive BP meter has been a matured healthcare device. This BP meter generally use oscillometric method with cuff and only provide intermittent measurement, thus it is not suitable for applying in long-term wearable device. Photoplethysmography (PPG) signals provide the information of terminal blood circulation as well as a simpler and noninvasive estimation of human biological signals. This paper aims for the better estimation of BP only by the PPG signals and an Artificial Neural Network (ANN). In this project, we will use MIMIC II database in PhysisoNet (https://www.physionet.org/) for BP estimation algorithm development and validation. PPG signals with corresponding BP values for different persons and different time instances are extracted from the MIMIC II database which includes ECG, BP, PPG which were recorded simultaneously with a sampling rate of 125Hz captured from vital sign monitoring in ICU. Machine learning approach (ANN in this project) will be applied to get BP estimation model. Ten features of PPG signals are extracted for the input vector for ANN, including the Max PPG amplitude, Total Area, Systolic Width, Diastolic Width, Systolic Area, Diastolic Area Systolic Width at 50%, Diastolic Width at 50% etc. Moreover, frequency-domain features of PPG signals will be used for ANN model construction. About 250,000 beat-by-beat PPG signals from 40 patients in MIMIC II database are used, 70% of which are used for training the ANN and 30% for external validation. The standards of British Hypertension Society (BHS) will be used to evaluate this proposed method. Grade A in this standard is our goal. PPG signal have been installed in wearable device for some application, e.g. heart rate estimation. The project develops a method for BP estimation only based on time and frequency domain features of PPG and demonstrate the feasibility of this method. The proposed method provides an opportunity for noninvasive and continuous BP measurement on a wearable device without extra sensing unit installation.
|Effective start/end date||8/1/18 → 7/1/19|
- Noninvasive Blood Pressure
- Artificial Neural Network (ANN)
- Time-frequency Analysis