Non-invasive cuff-less blood pressure estimation using LSTM-based recurrent neural networks

Modern technology significantly contributes to improving living conditions and decreasing disease prevalence. Wearable health tools have been at the forefront of key breakthroughs in the healthcare sector. The fact that these tools can be used in both normal activities and therapeutic applications has led to significant advancement in this field. Hypertension, characterized by periodic high blood pressure (BP), is the principal risk factor for Cardiovascular diseases (CVDs) that are among the leading causes of mortality. Hypertension often goes unnoticed by individuals, consequently, there is a pressing need for a monitoring device capable of continuously tracking blood pressure in various conditions of everyday life. Furthermore, nowadays to achieve continuous blood pressure monitoring is necessary to use invasive devices. However, these invasive methods pose limitations and additional challenges for patients. To address these concerns, the European SINTEC project, with the aim of developing a wearable health device (WHD) for continuous BP monitoring. The improvement of WHDs focuses on three key areas: enhancing sensor accuracy, implementing advanced machine learning algorithms, and facilitating seamless communication between subsystems and other devices. The accuracy of WHDs relies heavily on sensor technology. Advances in sensors have significantly improved the precision and reliability of measurements. Machine learning (ML) algorithms play a vital role in data analysis and interpretation in WHDs. These algorithms process the collected physiological data, identify patterns, and generate actionable insights. By leveraging large database, ML algorithms can continuously improve and adapt their performance. They enable personalized health recommendations and early detection of abnormal health conditions, empowering individuals to make informed decisions about their lifestyle and improve their health. WHDs now feature sophisticated communication skills that make it possible for them to be easily integrated with smartphones and PCs. This enables individuals to keep track of and share their health information with medical providers. The purposes of this thesis are to enhance the accuracy of BP monitoring devices and to eliminate noise and situational dependency. Due to the fact that ECG and PPG signals provide a safer and comfortable monitoring experience for people than invasive techniques, this is accomplished by using them to improve the accuracy of blood pressure estimation. While ECG measures the electrical activity of the heart using electrodes placed on the skin, PPG analyzes changes in blood volume using light sensors. The MIMIC-III database is used for training and testing the algorithm because consist of desired signals of diverse population. Signal cleaning techniques are employed, followed by feature extraction. Machine learning clustering algorithms are utilized to remove outliers from the extracted features. Finally, a novel neural network model incorporating the Long Short-Term Memory (LSTM) layer is developed to predict continuous systolic and diastolic blood pressure values. WHDs have witnessed remarkable advancements, driving the continuous improvement of non-invasive BP monitoring. These devices offer accurate and convenient solutions for individuals to monitor their cardiovascular health in real-time, both in daily life and medical settings. With the integration of advanced sensors, ML algorithms, and seamless communication capabilities, wearable health devices.