光體積變化描記圖之訊號處理方法比較與其應用於自主神經異常之篩檢工具設計

Abstract

研究背景 R-R間隔心律變異性 (R-R interval variation, RRIV)是最常用來測試心臟自主神經功能的檢查，它可以很可靠且準確地由心電圖electrocardiogram (ECG)的R波與R波之間的間隔推算得到。光體積變化描記圖Photoplethysmography (PPG)是一種可以由皮膚測得血氧飽和度的非侵入性檢測工具，一般而言，需要較長的檢查時間(5分鐘以上)，PPG測得的心跳結果才能和心電圖測得的結果接近，本研究的目的在經由訊號分析的方式改善PPG與ECG的一致性，進而發展一套簡單且非侵入性的自主神經功能篩檢工具。 研究設計 本研究招募20到69歲的成年人，收集他們同步的心電圖與光體積變化描記圖的結果。我們使用Pan-Tompkins演算法來分析心電圖中R波的訊號，使用三大類共十種濾波器來分析光體積變化描記圖當中心輸出波峰(systolic peak)的訊號， 第一類濾波器是數位濾波器-IIR無限脈衝響應，包括巴特沃斯(Butterworth)濾波器、貝賽爾(Bessel)濾波器、切比雪夫(Chebyshev)濾波器與橢圓(Elliptic) 濾波器；第二類是數位濾波器-FIR有限脈衝響應包括週期均值濾波法、移動均值濾波器與平滑演算法Savitzky–Golay；第三類是小波轉換包括連續小波轉換(continuous wavelet transform)、離散小波 Coiflet 3與離散小波 多貝西 (Daubechies 6) 以上總共十種各具特色的濾波器。 結果 本研究從2018年七月到2019年五月份總共檢測251名受試者，當中有7位受試者被排除，排除率是2.78%。在納入分析的244位受試者當中共有73(29.9%)位心臟自律神經功能異常，平均年齡是55.18±10.74。在三大類的濾波器當中IIR無限脈衝響應這組的表現最優異，其次是小波轉換，最後是有限脈衝響應。無限脈衝響應這組的四種濾波器表現在伯仲之間，當中，休息時以巴特沃斯(Butterworth)濾波器(0.51)、貝賽爾(Bessel)濾波器(0.51)絕對誤差最小，深呼吸時以切比雪夫(Chebyshev)濾波器(0.56)與橢圓(Elliptic) 濾波器(0.56)絕對誤差最小。巴特沃斯濾波器的接收者操作特征曲線(ROC curve)其曲線下方的面積高達0.982，顯示此濾波器擁有極佳的鑑別力。 結論 本研究顯示使用光體積變化描記圖的穿戴式裝置是個擁有高度潛力成為篩檢心臟自主神經功能異常的工具，無限脈衝響應這類型的濾波器能夠達到極優的敏感性同時兼顧特異性，未來還需要實際應用到社區作大量篩檢的研究來驗證這項工具的可信度。

Background and aim: The most widely used examination to test cardiac vagal tone is the R-R interval variation (RRIV) at rest and during deep breathing. RRIV can be reliably derived from electrocardiogram (ECG) data. Photoplethysmography (PPG) is a non-invasive method to record blood saturation from the skin. With PPG, pulse rate variation correlates with heart rate variation for longer periods of measurement. With the analyzed results from the study, we hope to construct a non-invasive and simple tool for the clinic diagnosis and screening of autonomic neuropathy. Design: We collected synchronize ECG and PPG signal from subjects between 20-69 years old. Pan-Tompkins algorithm was used for analyzing ECG signal. Ten filters, which fall into three main categories including infinite impulse response (IIR), finite impulse response (FIR), and wavelet transform (WT), were chosen to process the PPG signals. For the IIR filters, ButterWorth, Bessel, Chebyshev, and Elliptic filters were selected, and for the FIR filters, Savitzky–Golay, Average, and Period average filter included. Since WT includes discrete wavelet transform (DWT) and continuous wavelet transform (CWT), the selected WT filtering mechanisms with different mother wavelets like Daubechies and Coiflet were considered in this study. Results: From Jul. 2018 to May. 2019, a total number of 251 patients were included. 7 (2.78%) patients were excluded. Among 244 subjects, 73(29.9%) had abnormal cardiac autonomic function. The mean age was 55.18±10.74. Among three categories of filters, IIR filters had the best results followed by FIR and WT. Among IIR filters, Butter-Worth and Bassel filter had minima error (0.51%) at rest compared SSIV with RRIV. The error during deep breathing was almost the same between 4 IIR filters as following Butter-Worth 0.57%, Bassel 0.60%, Cheby1 0.56% and Ellip 0.56%. The AUC of ROC for Butter-Worth filter is 0.982 and the discrimination was outstanding. Conclusion: SSIV using PPG is a potential surrogate tool for massive screening of cardiac autonomic dysfunction. The Sensitivity and Specificity by IIR filters are outstanding discrimination and adequate for screening tool. Further studies to verify this tool in the community are expecting.