以光體積描述訊號偵測心房顫動之快速篩檢與長期監測應用

Abstract

心房顫動是一種因為心臟內產生節律訊號的功能異常，是最常見的心臟節律異常疾病，由於心房顫動亦是提高中風風險約五倍的危險因素，所以自動偵測心房顫動是一個重要的醫學議題，對於包含一般民眾的篩檢，以及疑似心房顫動的病患之長期監測而言。目前最有效的診斷為心電圖，然而，各種心電圖儀器都有一些缺點，例如，需要一邊肢體至少一個電極、或費用較高。另一方面，使用光體積描述訊號的偵測方式值得深入探討，光體積描述訊號是一種由血氧機可取得，可反應心律之信號。相較於心電圖較為簡便，因此，更適合長期監測病人生理訊號，並偵測像心房顫動等具有陣發性之心臟節律異常疾病。 然而，本論文中提到的以光體積描述訊號偵測心房顫動之現有相關著作，除了資料測量環境較標準化外，缺點多為較少考慮基線飄移，只考量單一的訊號參數序列等前處理問題，與較無探討特徵抽取的多寡與選取以及學習等後處理問題。因此，本論文提出一個以光體積描述訊號偵測心房顫動的架構，並且探討實際臨床應用上快速篩檢與長期監測的不同需求。在臺大醫院提供的實際臨床生理訊號資料中，本論文提出之架構之接收者操作特徵曲線的曲線下面積，靈敏度，特異度與準確度分別達到98.0%，95.4%，97.9%與97.3%，該數據較現有的光體積描述訊號偵測心房顫動之相關著作良好，且跟心電圖的心房顫動偵測架構準確度接近，並且在快速篩檢的應用上，可以將量測時間縮短至30秒並只增加少量的誤差，這說明本論文提出之光體積描述訊號架構有潛力被使用在快速篩檢與長期監測心房顫動。

Atrial Fibrillation (AF) is the most common and sustained type of cardiac arrhythmia. Since AF is a risk factor for stroke, automatic detection of AF is an important public health issue. Currently, the most useful and accurate tool for diagnosing AF is electrocardiography (EKG). However, EKG monitoring devices have their limitations or drawbacks. On the other hand, photoplethysmogram (PPG) is an alternative technique to obtain the heart rate information by pulse oximetry. Compared with EKG monitors, PPG devices are more convenient, making PPG promising in identifying paroxysmal AF. The aim of this thesis is to investigate the potential of analyzing PPG waveforms to identify patients with AF. The state-of-the-art PPG-based AF detection researches in this thesis have some limitations. In addition, there is still performance gap between related works and EKG-based algorithm. Therefore, we propose a PPG-based AF detection framework, including pre-processing, feature extraction, and SVM classification with GA-based optimization. The receiver operating characteristic curve (ROC) and statistical measures were applied to evaluate model performances. Furthermore, two clinical scenarios, long-term monitoring and fast screening were considered in the experiments. Among 673 patients’ signals recorded in clinic environments, we achieve ROC area under curve, sensitivity, specificity and accuracy of 0.980, 0.954, 0.979 and 0.973, respectively. And the record time can be shorten to 30 seconds with little performance degradation in fast screening scenario. The result suggests that the PPG-based AF detection algorithm is a promising pre-screening tool for AF and helps doctors monitoring patient with arrhythmia.