Heart health monitoring wearable device
Hin Wai Lui · 2018
In this thesis a heart health monitoring wearable device was developed to allow for convenient and frequent measurements of blood pressure (BP) and electrocardiogram (ECG), together with an MI classification system to automatically classify ECG records with MI. The MI classifier performs multiclass classification to discriminate ECG records of MI from healthy individuals, existing heart conditions, as well as records contaminated with noise. The method was tested on a database with MI ECG records. It was found that the addition of recurrent layer has improved classification sensitivity by 28.0% compared to convolutional neural network alone. Overall, it has achieved 92.4% sensitivity, 97.7% specificity, 97.2% positive predictive value, and 94.6% F1 score. Pulse transit time (PTT) has been a promising method to measure BP on wearable devices. However, to achieve acceptable accuracy, subject specific and frequent calibration is required. A novel calibration procedure was developed that can achieve accurate BP measurements with longer calibration intervals. The performance of the proposed procedure was tested on 10 subjects in a preliminary study, and on 33 subjects according to ESHIP and IEEE protocol. The results show that the new procedure has significantly improved the measurement precision over the single point calibration procedure, and has met the requirements of ESHIP. However, it was not sufficient to pass the test of induced BP changes from the IEEE protocol. Therefore, the proposed method might only be suitable when the subject is under a stable condition with recalibration once a week. Under a business model of device sales and premium subscription services, which users can receive personalised and data driven services from a nurse or a physician, the market opportunity in Hong Kong alone is expected to reach a total net profit of HK$82.4M in a 5 year period.