Development of a Wavelet-Based ECG Data Compression System with Real-Time Quality-on-Demand Function

[[abstract]]心電圖(ECG)提供一非侵入式檢測方式廣泛應用於心臟疾病之臨床診斷。心臟病患的照護與即時急救均需仰賴長期的心電圖紀錄與即時資料傳輸。以24小時紀錄的霍特(Holter)系統為例，每天約需要儲存45 MB的心電圖資料。為因應如此龐大資料的儲存與即時傳輸之需求，乃有多種ECG資料壓縮技術的發展。在失真度與壓縮率須妥協的失真壓縮法中，以SPIHT [8] 具最佳壓縮效能。SPIHT是基於小波轉換(wavelet-base)的一種壓縮編碼技術，其同時具最佳即時編碼能力與可硬體實現等優點。近來，我們結合一維離散週期性小波轉換(1-D NRDPWT)[九十一年度計劃研發成果NSC-91-2213-E-327-006]與可逆四捨五入線性轉換理論(RROLT)[九十四年度計劃研發成果NSC-94-2213-E-327-009]成功地發展出一新的心電圖資料壓縮技術 [21-23]。此技術應用三個策略以尋找最佳量化尺度：1)使用1-D NRDPWT有效地阻斷error propagation效應。2)應用RROLT理論取得具最小動態區間之小波轉換係數。3)利用RROLT理論正規化每一階小波轉換係數的重要性。基於此三策略，我們開發出兩種量化演算法；四捨五入法[21]和直接捨去法[23]，而以後者具最佳效果。直接捨去法與SPIHT方法的比較：在區間，PRD的平均改善率為18.65%；在12CR4≤≤20CR14≤≤區間，PRD的平均改善率為18.39%。總體的失真平均改善率達18.52%。由於心電圖訊號具非穩定特性，固定量化值的壓縮方式極易造成臨床資訊的嚴重流失。臨床診斷上是希望每一區段的心電圖訊號皆能維持固定的品質，此即心電圖訊號的品質保證機制。為滿足此一需求，心電圖資料壓縮系統必須同時具備即時壓縮能力和隨選品質(Quality on Demand)功能，此即本計劃之研究目的。本計劃擬以兩年兩階段方式完成雛型系統的研發。第一階段改進量化尺度之規劃，使得失真度對量化控制參數之曲線更趨近線性，但不犧牲目前的壓縮性能。第二階段利用此線性特性有效地降低誤差控制迴圈的計算複雜度和迴圈重複次數，使得品質控制可於千分之一秒內完成。本計劃之初期研究成果已為IEEE BMEI 2008 (Hainan, China)國際研討會所接受[25]。[[abstract]]Electrocardiogram (ECG) signal analysis is a non-invasive modality widely used for heart disease diagnosis and ambulatory monitoring. Motivated by the demands of long time recording and real-time data transmission, many loosy ECG data compression methods under the crucial requirements of preserving clinical information have been proposed. In these methods, the SPIHT scheme [8] with excellent compression performance and realizability is the best one. Recently, a new ECG data compression technique based on the 1-D non-recursive discrete periodized wavelet transform (1-D NRDPWT) and the reversible round-off linear transformation (RROLT) theorem [21-23] was proposed. The 1-D NRDPWT developed in the project of NSC-91-2213-E-327-006 is a non-recursive DPWT that can effectively resist error propagation effect. The RROLT theorem developed in the project of NSC-94-2213-E- 327-009 can perform the 1-D NRDPWT with minimal word length guaranteed and normalize the significance of each band wavelet coefficients. With the three facilities, two quantization algorithms, round-off method [22] and truncation method [23], have been presented. The later is better than the former. As compared with the SPIHT scheme, the truncation method can improve the average PRD by 18.65% and 18.39% for 12CR4≤≤ and 20CR14≤≤, respectively. The total improvement of the average PRD is about 18.52%. In lossy ECG data compression, quality control of reconstructed data is an important issue that requires retrieved ECG signals with desired quality guaranteed. Data compression methods based on fixed bit rate or fixed quantization scale, generally, can not satisfy the requirement due to the non-stationary property of ECG signal. A quality-on-demand (QOD) system requires an error control loop to limit the distortion of every segment data within a small region. The main goal of this project is to develop a real-time wavelet-based ECG data compression system with QOD function. Two stages are planned for performing this project in two years. The first stage will improve the linearity of distortion-QF curve without the cost of compression performance where QF is an adjustable parameter used for compression performance control. The second stage utilizing the linearity will develop an efficient error control loop that guarantees the desired quality can be obtained within 1 ms. A primary work of this project is going to be published in the IEEE international conference on BMEI, 2008 (Hainan, China) [25].[[note]]NSC97-2221-E327-02