On Designing an ECG-based Intelligent System: Utilizing the Heart’s Electrical Activity to Recognize Humans and Detect Arrhythmia

Abstract

The electrocardiogram (ECG) signal is the bioelectrical signal that reflects the heart's activity. It has been extensively used as a diagnostic tool since it holds information about the cardiac health condition. However, recent researches have shown that it exhibits an inter-subject variability property. Therefore, it can be used as a biometric-based modality for either identification or verification purposes. Nevertheless, some of the challenges are faced while employing such a signal. For instance, ECG signal is prone to noise, accordingly, noise filters should be designed to remove the noise while keeping the signal properties. Moreover, factors such as medications, health condition, and emotional state can affect the intra-subject variability of the signal. Therefore, experiments that are performed should consider these factors to produce a robust system. In this thesis, we utilize the ECG signals to propose three approaches: one clinical approach, and two biometric approaches. In the first approach, we automatically detect cardiac arrhythmia, a disease when the heart fails to contract at the normal rhythm, using the signal textures. These texture operators include both temporal and spectro-temporal features. The temporal texture includes the one-dimensional local binary pattern (1D LBP), while the spectro-temporal textures include the investigation of the texture of short-time Fourier transform (STFT) and generalized Morse wavelet transform (CWT). Different classifiers as well as different ECG lead configurations were studied. Classification accuracies of 92.97%, 99.24%, and 99.81% on MITDB were achieved using the 1D LBP, CWT, and STFT textures respectively. On the other hand, the ECG signal showed some advantages over other biometric modalities such as face and iris. These advantages incorporate the embedded liveness detection, the difficulty to counterfeit when compared to other traits, and the hybrid information it holds about the subject identity as well as the health, mental, and emotional conditions. In this work, we propose two biometric approaches. The first approach utilizes the spectrotemporal properties of the signal for both subject identification and validation. A two-dimensional convolutional neural network (2D CNN) was employed to extract the high-level features to distinguish subjects. Both the STFT and CWT were investigated as features. Validating on eight databases, an overall identification rate of 97.85%, equal error rate of 0.0268, and area under curve of 0.99% were achieved for STFT features. The second approach utilizes the cyclostationary properties of the signal since the ECG signal is non-stationary. The superiority of this approach is that it follows the blind-segmentation where no ECG wave points are detected, as well as being more robust to the noise in the ECG signal. The spectral function of the signal is extracted and used as features to feed a 2D CNN. Seven databases were used to validate the robustness and generalization of the approach, where an average identification rate ...