Cardiovascular Signal Processing: State of the Art and Algorithms

Abstract

The emergence of Artificial Intelligence (AI) has brought many advancements in biomedical signal processing and analysis. It has opened the way for having efficient systems in the diagnosis and treatment of diseases such as Cardiovascular (CV) disorder. CV disorder is one of the critical health problems causing death to lots of peoples globally. Electrocardiogram (ECG) signal is the signal taken from the human body to diagnosis the status of CV and heart conditions. Earlier to the introduction of computers, the diagnosis of heart conditions was made by experts manually and that caused various mistakes. Currently, the usage of advancing signal processing devices help to reduce those errors and enables to develop effective signal detection and parameter estimation algorithms that are useful to analyze the parameters of ECG signals. Which intern supports to decide if the person is in critical condition and take an appropriate action. In this work, we analyze the performances of classical techniques and machine learning algorithms for ECG based CV parameters estimation. For this, first an in-depth review is done for both classical techniques and machine learning algorithms. Specifically, the benefits and challenges of machine learning and deep-learning algorithms for CV signal processing and parameter estimation is discussed. Then, we evaluate the performances of both classical (Kalman Filtering) and machine learning algorithms. The machine learning based algorithms are modeled with Butterworth low pass filter, wavelet transform and linear regression for parameter estimation. Besides, we propose an algorithm that combines adaptive Kalman filter (AKF) and discrete wavelet transform (DWT). In this algorithm, the ECG signal is filtered using AKF. Then, segmentation is performed and features are extracted by using DWT. Numerical simulation is done to validate the performances of these algorithms. The results show that at ${20}{\%}$ false positive rate, the detection performance of Kalman filtering, the proposed algorithm and machine learning algorithm are ${83}{\%}$, ${94}{\%}$ and ${97}{\%}$, respectively. That shows the proposed algorithm gives better performance than classical Kalman filtering and has nearly the same performance with machine learning algorithms.