In this study, we will be using 3-lead ECG electrodes to collect ECG Signals. This approach simplifies the process of identifying the correct waves and eliminates the need for precise placement in specific locations. Our main objective is to clean the ECG data collected by the hardware team which will be helpful in the feature extraction to accurately estimate the BP. The proposed system consists of the following steps: 1. The ECG Signal data: We will collect and store the ECG Signal data, which will serve as the fundamental inputs for our model. 2. Preprocessing the ECG Signal: The ECG Signal will undergo preprocessing to ensure they are suitable for model training, validation, and testing. 3. We will create a Deep Neural Network (DNN) model that can denoise the ECG Signal and help in extracting useful features from the preprocessed signals during the BP Estimation. This model will be trained to estimate denoised signals concurrently. 4. Calibration-free model training, validation, and testing: We will train, validate, and test the model without the need for calibration, making the process more efficient and user-friendly.

In the recent article on Nature Reviews[2] of Prof. Vincent Mok (Principal Investigator), he pointed out that small vessel diseases can play a critical role in delayed-onset poststroke dementia. Therefore, people who have survived stroke are at risk of developing dementia beyond the early poststroke period. Other studies suggest that midlife hypertension leads to increased cognitive decline, and could thus contribute to the development of vascular dementia, and even Alzheimer’s disease, in late life[3-4]. A longitudinal study over a 32-year period demonstrates that those who developed dementia, particularly vascular dementia, had a greater increase in systolic blood pressure (BP) from midlife to late life, followed by a greater decrease over the late life, and such association was significantly reduced in those taking anti-hypertensive medications [5].

This study includes monitoring of non-invasive continuous BP. ECG, PPG and pressure pulse wave signals. These signals will be acquired with one or more of the following sensors positioned on the patients. Non-invasive continuous measurement, ECG electrodes on the left and right lower arm, BP on the left and right finger, PPG on the left index fingertips, continuous BP waveform on the left hand side and continuous pressure pulse wave signal on the left and right arm wrist

In this study, we will develop a novel, efficient, generalized, and feature engineering exempted data-driven DL approach for estimation of BP employing a single channel PPG signal recorded from diverse subjects with and without CVD complications. A modified long-term recurrent convolutional network (LRCN) based DL framework is proposed that combines the strengths of CNN and bidirectional LSTM (BiLSTM) to simultaneously infer systolic blood pressure (SBP), and diastolic blood pressure (DBP) by exploiting a single sensor unit data (e.g., PPG sensor data). The model benefits from data-driven feature extraction by leveraging the joint framework of CNN-BiLSTM, resulting in an efficient and precise solution for wearable healthcare IoT applications. The proposed LRCN framework achieves several goals and contributes to the advancement of sciences: 1) The proposed model predicts SBP and DBP simultaneously, offering a cost-effective solution with minimal sensor units. 2) The development of a sophisticated BP framework eradicates the requirement for individual model training. 4) The LRCN model with improved average MAE and SD for BP estimation on a larger population can be a tremendous achievement as compared with recently reported research. 4) Continuous monitoring of BP enhances the potential for initial detection of human health conditions and leads to improved outcomes.

Clinical Partner: Prof. Bryan YAN (at HKSTP) We employ a 16 channel data collection modal to gather data simultaneously and continously for the analysis and mapping of PPG, pulse transit time and BP

We have developed electrode materials on a flexible substrate using a new solution method, which are almost identical to traditional rigid silver/silver chloride electrodes. Measurements show that the new materials have the same XRD patterns and surface impedance as the original materials. We aim to develop these new electrodes into usable bioelectrode materials that can produce high-definition flexible bioelectrodes. Our new materials and designs have good biocompatibility and compatibility, and are less likely to cause skin allergies during long-term wear. We will further test their stability and comfort on the skin surface and ultimately analyze their market potential.