| 摘要: | 心肌梗塞（又稱心臟病）在過去 10 年內持續位居全球十大死因之一。根據世
界衛生組織（WHO）2023年的統計資料，每年因心血管疾病死亡的人數高達1,800
萬人，占全球總死亡人數的32%。同時，心肌梗塞患者依阻塞程度不同，常伴隨多
種併發症，例如心律不整、心包膜炎，嚴重者甚至可能導致心臟衰竭或心臟破裂而
死亡。雖然這些病徵具有高度危險性，但透過及早治療可顯著降低風險。目前常見
的治療方式包括經皮冠狀動脈介入治療（Percutaneous Coronary Intervention, PCI），
利用支架置入或血栓溶劑等方法，然而在治療前如何精準檢測心肌梗塞區域仍是臨
床上亟需解決的問題。
現有檢測方式如心臟超音波、心導管及心臟 MRI 雖廣泛應用，但存在準確性
不足、成本高、耗時長、侵入性強及輻射暴露等缺點，且無法進行長期監測，限制
了急性心肌梗塞的篩檢應用。標準12 導程心電圖能夠粗略描述病人之心電生理之
情況，若能透過心電圖針對缺血區域定位即可以達成上述之理念。當心肌缺血時，
由於心肌細胞不再具有傳導之能力，在心臟電位上會發生激發電位降低、靜止電位
上升等情況，進而導致臨床上利用心電圖觀察缺血問題之特徵常包含ST間期電位
上升及下降、T波段反轉等行為。
本研究應用數位孿生心臟模型產生不同心軸角度以及不同缺血區域、等級之標
準12導程心電圖資料庫，並且參考臨床特徵後針對每一導程擷取ST間期之5種特
徵再以稀疏式編碼分類器對兩開源資料庫：PTB、PTB-XL中總計348位含有心肌
梗塞病徵之資料分類。透過上述方式在區域定位上獲得0.75之準確度，並且在偵測
有無心肌梗塞之應用上獲得高達0.99之敏感度。透過此技術，本研究能以非侵入式
的檢測方式對病人進行精準檢測同時能利用數位資料之特性達成遠端醫療之理念。;Myocardial infarction (MI), commonly known as heart disease, has remained one of the
top ten causes of death globally over the past decade. According to World Health
Organization (WHO) statistics from 2023, cardiovascular diseases account for
approximately 18 million deaths annually, representing 32% of all global mortality.
Depending on the degree of coronary artery blockage, MI patients may experience severe
complications such as arrhythmias, pericarditis, heart failure, or cardiac rupture. Although
early treatment significantly reduces these risks, accurate localization of the infarcted
region prior to intervention remains a major clinical challenge.
Current diagnostic tools, including echocardiography, cardiac catheterization, and
cardiac MRI, are widely used but suffer from limitations such as high cost, invasiveness,
time consumption, and insufficient accuracy, making them less suitable for long-term or
large-scale screening. The standard 12-lead electrocardiogram (ECG), while non-invasive
and widely available, only provides a coarse overview of cardiac electrophysiology.
However, if ischemic region localization can be achieved through ECG, it would enable
more practical and scalable diagnostics.
This study presents a novel approach that utilizes a digital twin heart model to generate
a synthetic 12-lead ECG dataset under varying cardiac axis angles, ischemic locations, and
severity levels. Based on clinical patterns, five ST-segment features were extracted from
each lead and fed into a sparse coding–based classifier. The model was evaluated on two
open-source datasets: PTB and PTB-XL, comprising a total of 348 MI cases. The proposed
method achieved a localization accuracy of 0.75 and a sensitivity of 0.99 for infarction
detection. These results demonstrate the feasibility of using non-invasive ECG data for
precise infarct localization and suggest the potential for remote monitoring and
telemedicine applications through digital modeling. |
