| dc.description.abstract | Cardiovascular disease (CVD) is the global number one cause of death. Therefore, there is an acute need for constantly monitoring cardiac conditions and/or cardiac monitoring for extended periods. The current clinical Electrocardiogram (ECG) recording systems require precise placement of electrodes on the patient’s body, often performed by trained medical professionals. These systems also have long wires that require repeated disinfection and can be easily tangled and interfered with clothing and garment. These limitations have severely restricted the possible application scenarios of ECG systems. To overcome these limitations, there is a need for wearable ECG devices with minimal wires to detect possible cardiac abnormalities with minimal intervention from healthcare professionals.
Previous research on this topic has focused on extracting cardiac information from the body surface by investigating various electrode placements and developing ECG processing algorithms. Building on these studies, it is possible to develop devices and algorithms that can extract ECG-related information without the need for precise electrode placements on the body's surface. The present thesis aims to extract ECG-based cardiac information using signals recorded from the upper arm.
Far-field ECG is prone to contamination by artifacts such as Electromyogram (EMG), which greatly reduces its clinical value. The current study examines how various state-of-the-art heartbeat detection algorithms perform in four levels of simulated EMG artifacts. The simulated EMG was added to Lead II from two different datasets: the MIT-BIH arrhythmia dataset (Dataset 1) and data we collected from 20 healthy participants (Dataset 2). Results show that Stationary Wavelet Transform (SWT) provided the most robust features against EMG intensity level increment among various algorithms. The next step involved recording bio-potential signals using a high-density bio-potential amplification system attached to the upper arm. The system used three high-density electrodes, each with 64 channels, in addition to the standard Lead II. Twenty participants, reported healthy, were asked to perform two tasks: Rest and Elbow Flexion (EF): holding three weights (C1: 1.2 kg, C2: 2.2 kg, and C3: 3.6 kg). The tasks were repeated 2 and 3 times, respectively. Firstly, I identified optimal electrode locations on the upper arm for each task. I then generated a synthesized ECG using the selected electrodes with generalized weights over subjects and trials. Considering the robustness of SWT to EMG intensity level increment, I next focused on optimizing SWT by addressing two of its drawbacks: introducing phase shift and the requirement of a pre-defined mother wavelet. Regarding the first drawback, zero-phase wavelet (Zephlet) was implemented to replace SWT filters with zero-phase filters for the matter of feature extraction from the synthesized ECG. Next, I incorporated the synchronized extracted features with a Multiagent Detection Scheme (MDS) for the means of heartbeat detection. The F1-score for the heartbeat detection was 0.94 ± 0.16, 0.86 ± 0.22, 0.79 ± 0.26, and 0.67 ± 0.31 for Rest and EF with three different levels of muscle contraction (C1 to C3), respectively. Changing the acceptable distance between the detected and actual heartbeats from 50 ms to 20 ms, the F1-score changed to 0.81 ± 0.20, 0.66 ± 0.26, 0.57 ± 0.26, and 0.44 ± 0.26 for Rest and C1 to C3, respectively. Regarding the second drawback, Lattice parametrization was used to optimize the mother wavelet for the means of PQRST delineation. The mother wavelet was generalized over subjects, trials, and tasks. The Pearson’s Correlation Coefficient (CC) between the averaged delineated PQRST from analyzing feature and the averaged PQRST from Lead II using this generalized mother wavelet was 0.88 ± 0.05, 0.85 ± 0.08, 0.83± 0.11, and 0.81 ± 0.12 for Rest and C1-C3, respectively.
This thesis makes several contributions to the current literature. It introduces locations on the upper arm that can be used to place sensors in a wearable to capture cardiac activity with robustness across intra-subject, inter-subject and inter-contraction variabilities. It also identifies a robust method against noise increment for heartbeat detection. Zephlet was implemented for the first time that can replace SWT in many applications in which there is a need for synchrony with respect to the original signal or among components. And finally, this thesis introduces a generalized mother wavelet that can be used to extract PQRST and enhance SNR in many applications, such as ECG waveform extraction, arrhythmia detection, and denoising. | en |
