Development of Some Efficient Schemes for ECG Signal Quality Assessment and Processing

Abstract

ElCTROCARDIOGRAM (ECG) is a widely utilized tool for the diagnosis of cardiac ailments. However, considering the rapid growth of population as well as poor doctor to patient ratio, the computer-based automated monitoring of ECG signal is efficient as it provides fast processing and precise long term monitoring with less human effort. The main objective of this research work is to improve the performance of a computer-based automated ECG monitoring system incorporating efficient schemes for ECG signal quality assessment and processing, which can enhance the reliability of the medical diagnosis. Some of the existing literature on signal quality assessment (SQA) discuss ECG SQA schemes employing QRS complex morphology and R-peak information based features. But the performance of such techniques degrades due to improper detection of R-peaks. This research attempts to develop an efficient wavelet sub-bands features based ECG SQA scheme. Various wavelet sub-band level signal features are applied to machine learning-based classifier for estimating the ECG signal quality. Further, the filtering of ECG records is essential for precise identification and detection of local components when a considerable amount of noise in ECG records affects the low amplitude P and T waves significantly. Hence this research proposes an efficient ECG denoising methodology based on both empirical mode decomposition (EMD) and adaptive switching mean filter (ASMF) approach. Unlike conventional EMD based techniques, the proposed method reduces the effect of noises simultaneously preserving the high-frequency QRS complex information. For typical low-frequency noises, this method does not meet the achievable performance. Hence, this research is motivated towards developing ECG signal enhancement scheme in a dictionary learning (DL) based sparse representation framework. The proposed method completely enhances the ECG signal quality by removing both low and high-frequency noises. Continuous monitoring of ECG records for better medical diagnosis in telemetry needs a large memory space for storage and also it requires more transmission power. A low power based solution utilising for ECG compression based on beat type dictionary based compressed sensing (CS) scheme is proposed which offers high-quality signal recovery avoiding the training stage of individual ECG record. Effective detection of R-peaks in ECG plays a vital role in the extraction of clinical features in a computer-based automated system. Hence, a wavelet-based R-peak detection scheme using the Hilbert transform is proposed. The efficacy of all the proposed approaches is evaluated through MATLAB simulation based studies by comparing with the existing schemes considering ECG records of standard databases.