Date of Award
Doctor of Philosophy (PhD)
Stan A. Napper
Performance improvement in computerized Electrocardiogram (ECG) classification is vital to improve reliability in this life-saving technology. The non-linearly overlapping nature of the ECG classification task prevents the statistical and the syntactic procedures from reaching the maximum performance. A new approach, a neural network-based classification scheme, has been implemented in clinical ECG problems with much success. The focus, however, has been on narrow clinical problem domains and the implementations lacked engineering precision. An optimal utilization of frequency information was missing. This dissertation attempts to improve the accuracy of neural network-based single-lead (lead-II) ECG beat and rhythm classification. A bottom-up approach defined in terms of perfecting individual sub-systems to improve the over all system performance is used. Sub-systems include pre-processing, QRS detection and fiducial point estimations, feature calculations, and pattern classification. Inaccuracies in time-domain fiducial point estimations are overcome with the derivation of features in the frequency domain. Feature extraction in frequency domain is based on a spectral estimation technique (combination of simulation and subtraction of a normal beat). Auto-regressive spectral estimation methods yield a highly sensitive spectrum, providing several local features with information on beat classes like flutter, fibrillation, and noise. A total of 27 features, including 16 in time domain and 11 in frequency domain are calculated. The entire data and problem are divided into four major groups, each group with inter-related beat classes. Classification of each group into related sub-classes is performed using smaller feed-forward neural networks. Input feature sub-set and the structure of each network are optimized using an iterative process. Optimal implementations of feed-forward neural networks provide high accuracy in beat classification. Associated neural networks are used for the more deterministic rhythm-classification task. An accuracy of more than 85% is achieved for all 13 classes included in this study. The system shows a graceful degradation in performance with increasing noise, as a result of the noise consideration in the design of every sub-system. Results indicate a neural network-based bottom-up design of single-lead ECG classification is able to provide very high accuracy, even in the presence of noise, flutter, and fibrillation.
Thiagarajan, Srikanth, "" (2000). Dissertation. 151.