Blood Pressure measurement using a smartwatch

Tran Minh Quang
Le Nguyen Minh Huy
Ngo Quang Trung
Nguyen Thanh Thuan

Cardiovascular disease (CVD) is the leading cause of death globally, accounting for over 30% of deaths. Specifically, in Vietnam, cardiovascular disease was responsible for 31% of all deaths in 2016, equivalent to more than 170,000 deaths. Constantly monitoring the current blood pressure (BP) is vital for CVD patients since, based on the BP, the doctor and patient can take appropriate action to prevent the BP from a dangerous stage. However, the BP device and monitor methods at the moment are inconvenient, which usually use an upper arm monitor device and take about 20 to 30 seconds to get the result. Thus, the need for a more compact and convenient method for estimating BP is becoming popular in recent years. 

The project aims to implement a blood pressure measurement feature on wearable devices, which predicts systolic and diastolic blood pressure with high accuracy even with the data obtained from wearables devices realistic. Particularly, the team will collect raw PPG from public datasets such as the vital sign Queensland dataset, pre-processing the digital signals and conducting feature engineering steps to select a set of potential features. It is observed that the dataset with raw PPG signals easily contains many noises such as breathing and motion artifacts. Thus, it is essential to improve the signal quality over noise by doing wavelet transformation, average filtering and normalizing the single beat PPG waveform. Also, outliers’ data need to be removed to prevent a negative effect on the final results. Then, we apply machine learning techniques to develop an efficient algorithm estimating continuous BP. Some experiments with different machine learning model with from traditional machine learning (SVM, Random forest...) to deep learning model (LSTM, MLP) will also be performed for comparison purposes and the best one with the smallest mean standard deviation error will be chosen.