Time-Domain and Transform-Domain Compression Algorithms for ECG Signals with Irregular Periods

• Main Authors: Hsiao-Hsuan Chou, 周曉璇
• Other Authors: Te-Son Kuo
• Format: Others
• Language: en_US
• Published: 2006
• Online Access: http://ndltd.ncl.edu.tw/handle/65093338276150914793

博士 === 國立臺灣大學 === 電機工程學研究所 === 94 === Because modern Electrocardiogram (ECG) monitoring devices generate vast amounts of data and require huge storage capacity, many ECG compression methods have been proposed to process, transmit, and store the data efficiently. Most of the related papers showed fair ECG compression performances for regular ECG cases. However, their compression performance dropped in irregular ECG waveforms. In fact, the abnormal ECG signals have more clinic significance. In this dissertation, we propose improved time-domain and transform-domain compression algorithms separately for ECG signals with irregular periods. For the time domain, a novel and rapid ECG signal compression algorithm with less error for non-uniform sampling is proposed. It meets the real-time requirements for clinical applications. Moreover, the compression performance is stable even for abnormal ECG signals. A criterion called the Sum Squared Difference (SSD) is first defined as an error test equation. The algorithm using SSD to calculate error tolerance is applied to the records in the MIT-BIH 11-bit resolution database that was based on a 360 Hz sampling rate. It belongs to the threshold-limited algorithm such as the popular Fan algorithm but outperforms the Scan-Along Polygonal Approximation (SAPA), the Fan, and the Maximum Enclosed Area (MEA) algorithms in Sample Compression Ratio (SCR) and the Percent Root mean squared Difference (PRD). In addition, it maintains more clinical features of the ECG signals. For the transform domain, this dissertation presents an effective and efficient algorithm for compressing ECG signals by exploiting their inter- and intra-beat correlations. To better reveal the correlation structure, the ECG signals are converted into a proper 2-D array. This involves a few steps including QRS detection and alignment, period sorting, and length equalization. Of all the steps, period sorting has been first proposed by us as a novel and powerful method to reduce period differences among heartbeats effectively. Then the state-of-the-art JPEG2000 is selected for its high efficiency and flexibility. In this way, the proposed algorithm is shown to outperform existing methods in the literature by simultaneously achieving high Compression Ratio (CR) and low PRD. Furthermore, because the proposed period sorting method rearranges the detected heartbeats into an orderly array that is easier to compress, this algorithm is insensitive to irregular ECG periods. This is a significant improvement over existing 2-D ECG compression methods. This algorithm can be combined with other algorithms or codecs to improve their efficiency.