Regression algorithms for estimation of cuffless blood pressure estimation

Hypertension is a significant risk factor for cardiovascular diseases (CVDs), which are a leading cause of death around the world. Continuous monitoring of blood pressure (BP) is a proven tool to support patient care and when used in combination with other vital parameters such as heart rate, breath frequency, and physical activity, it can be highly effective in the prevention of CVDs. However, invasive methods are currently the most reliable way to continuously monitor BP, despite the potential for discomfort and damage to the patient. Non-invasive techniques are currently not considered optimal for continuously monitoring BP trends, as they can only return BP values every few minutes. In this thesis work, cuff-less estimation of continuous BP through Pulse Transit Time (PTT) and Heart Rate (HR) using boosting regression techniques is investigated. This method achieves non-invasive estimation of BP with an acceptable low error, according to the AAMI guideline. The study introduces several methods, including the use of electrocardiographic (ECG), photoplethysmographic (PPG) signals and ABP (Arterial blood pressure) signals extrapolated from the MIMIC III online database, and the implementation of preprocessing of the ECG and PPG signals, and the research and processing of the features related to them in order to continuously monitor BP in a non-invasive way, exploiting boosting regression techniques. Recent studies have demonstrated that HR and PTT can be linearly combined to obtain BP values, manipulation of these two parameters is key to non-invasively estimating reliable BP values. The goal of this research is to examine the use of boosting regression techniques to estimate continuous blood pressure in a non-invasive way that is equally depend- able with respect to current methods and easy for the patient to carry out, The user experience of this technology allows for the patient to easily and comfortably measure their blood pressure at various times throughout the day, regardless of their location, even if they are not in a clinical setting. This proposed approach can be viewed as the initial step towards the incorporation of these algorithms into wearable devices, particularly those developed for the SINTEC project.