多項性生理訊號實現急性中風病患預後與嚴重度偵測

Abstract

中風是造成死亡與失能的主要原因，心房顫動更是提高中風風險約五倍的危險因素，若能早期預測有心房顫動的中風病患的預後，對於他們的治療會是有幫助的。然而，目前的診斷設備，例如電腦斷層掃描與核磁共振，其缺點為昂貴、不可攜、可能導致副作用；若是根據無心房顫動的中風病患的預後方式採用多項性生理訊號如心電圖、動脈血壓、光體積描述訊號來早期預測中風病患的預後，會發現心電圖和光體積描述訊號並不是很準確，但是以動脈血壓作為量測方式的話，雖然較為準確但其缺點是侵入式量測，不易於每位病患上取得。 因此，本論文提出以多項性生理訊號，包括：光體積描述訊號和脈波傳遞時間來早期預測有心房顫動的中風病患的預後。透過多項性生理訊號分析架構，本論文預測心房顫動中風預後的準確率達到84.2%，該準確率較單一種生理訊號預測準確率佳，這意味著本論文提出的多項性生理訊號分析架構有潛力用於預測有心房顫動的中風病患的預後。 此外中風是一個動態的過程，他有可能隨著時間惡化，假如我們想要即時監控中風病患當下的嚴重程度變化，我們必須要有另外一個方法和指標來判斷它，目前醫療人員在判定中風病患的嚴重程度是透過例如語言、行動能力這些個體功能來評估中風的嚴重程度，但這些量表的數值在中風病人的個體功能產生劇烈的變化之前是無法反映出來的，所以我們提出以生理訊號來做進一步的分析。本論文提出基於多項性生理訊號的中風嚴重程度惡化偵測架構，其接收者操作特徵曲線的曲線下面積達到94.7%，這意味著本論文提出的多項性生理訊號的中風嚴重程度惡化偵測架構有潛力用於中風嚴重程度監控。

Stroke is a leading cause of death and disability. Atrial fibrillation is a risk factor that increases the risk of stroke by about five times. If the functional outcomes of stroke patients with atrial fibrillation can be predicted early, it will be helpful for their treatment. However, current diagnostic devices, such as computed tomography (CT) and magnetic resonance imaging (MRI), have the disadvantage of being expensive, unpotable, and potentially causing side effects; and if we use the multi-modal analysis methodology based on physiological signals, using in non-atrial fibrillation (non-AF) to predict the outcome prediction of stroke patients, we find that the multiple physiological signals such as electrocardiogram (EKG) and photoplethysmogram (PPG) are not very accurate, but with arterial blood pressure (ABP), as a measurement method, although it is more accurate, its shortcoming is invasive measurement, not easy to obtained on every patient. Therefore, this paper proposes a multi-modal analysis methodology to early predict an atrial fibrillation (AF) stroke patient’s functional outcome based on physiological signals, including: PPG and pulse transit time (PTT). Through the multi-modal analysis framework, the accuracy of the outcome prediction of AF stroke is 84.2%, which is better than the single modal. This means that the multi-modal analysis framework proposed in this paper has potential to predict the outcome of stroke patients with atrial fibrillation. In addition, stroke is a dynamic process, and it may deteriorate over time. If we want to monitor the current severity of stroke patients, we must have another method and indicator to judge it. The method of judging a stroke patient’s severity is to assess the severity of stroke through individual functions such as language and mobility. However, the values of these scales are not reflected obviously before the dramatic changes in the individual function of stroke patients, so we propose a physiological signals framework for stroke patients’ severity monitoring. This paper proposes a detection system for the severity of stroke severity based on physiological signals. The area under the curve of the receiver's operating characteristic curve (ROC) is 94.7%, which means that the multi-modal physiological signal proposed in this paper has a severity of deterioration detection architecture has the potential to be used for stroke severity monitoring.