結合體表心電圖及深度學習發展心血管疾病患者之泛化性風險預測模型與個人化數位孿生心臟;Utilizing Body Surface Potentials and Deep Learning to Develop Generalizable Risk Prediction Models and Personalized Digital Twin Hearts for Cardiovascular Disease Patients

NCUIR > college of Health Sciences and Technology > Institute of Biomedical Engineering > Electronic Thesis & Dissertation > Item 987654321/94788

Please use this identifier to cite or link to this item: http://ir.lib.ncu.edu.tw/handle/987654321/94788

Title:	結合體表心電圖及深度學習發展心血管疾病患者之泛化性風險預測模型與個人化數位孿生心臟;Utilizing Body Surface Potentials and Deep Learning to Develop Generalizable Risk Prediction Models and Personalized Digital Twin Hearts for Cardiovascular Disease Patients
Authors:	彭鈺琪;Peng, Yu-Qi
Contributors:	生物醫學工程研究所
Keywords:	心血管疾病;深度學習;數位醫療;心臟衰竭;心房顫動;數位孿生心臟;心電圖逆問題;心電圖成像;cardiovascular diseases;deep learning;digital medicine;heart failure;atrial fibrillation;digital twin heart;inverse probelm of electrocardiography;ECG imaging
Date:	2024-07-26
Issue Date:	2024-10-09 15:30:18 (UTC+8)
Publisher:	國立中央大學
Abstract:	根據世界衛生組織(WHO)公布的全球前十大死因調查中，心血管疾病包含缺血性心臟疾病和中風佔據該排行前二名將近20年，且每年有1790萬人死於心血管疾病，相當於三分之一的死亡人數。了解心臟的電生理活動有助於心臟疾病的診斷與治療，心電圖紀錄了心臟放電活動訊號，為臨床上非侵入式檢測心臟疾病的利器，配合深度學習強大的運算能力，有望協助臨床上的疾病診斷，達成數位醫療和即時診斷的目標。　　本研究使用體表心電圖發展多項研究主題，囊括應用深度學習於心血管疾病辨識和風險評估的研究以及建立個人化數位心臟模型，藉以剖析深度學習於臨床應用的可能性，透過不同面向研究模型的穩定性、跨裝置與跨資料庫之泛化能力和真實應用情境的效能。此外，使用高密度的體表心電圖與心電圖成像技術，運用非侵入的方法建立個人化的心臟模型，以期協助個人化的精準醫療決策。研究主題包含四大項：(1)單導心電圖於辨識低左心室射血分率之心臟衰竭，使用穿戴式裝置心電圖配合居家量測，檢測出潛在心臟功能與構造異常患者，提醒盡早就醫檢查；(2)從竇性心律12導心電圖中找出潛在的陣發型心房顫動，並應用於與心房顫動息息相關的中風患者之風險評估；(3)針對急性心肌梗塞患者的預後死亡風險預測與長期死亡風險分層能力；(4)使用心電圖成像技術建立個人化數位孿生心臟，多數心臟疾病源自於心臟構造異常與心肌細胞基質改變，進而影響心肌細胞的傳導，從CT影像與體表心電圖重構心臟電位，藉以分析異常的心臟放電情形與傳導途徑。　　上述主題涵蓋不同使用場合與情景的應用，從醫院臨床資料到居家個人使用、適用大眾群體到專注於特定患者群體的模型、群體適用之模型到專注於個人化的孿生心臟模型，發展從群體到個體層面的各式疾病診斷協助工具。 ;Cardiovascular diseases, including ischemic heart disease and stroke, have been the leading causes of death according to the World Health Organization (WHO) over the past 20 years. Understanding electrophysiology is crucial for comprehending the pathogenesis and diagnosis of cardiovascular diseases. An electrocardiogram (ECG) is a noninvasive tool that records the electrical cardiac activity through the body surface. With the strong computational capabilities of deep learning (DL) models, its application in clinical diagnosis is promising. This aligns with the goals of digital medicine and instant diagnosis. This study covers several topics that utilize ECGs and DL models for cardiovascular disease identification and risk stratification, aiming to analyze the feasibility and generalizability of DL models in real-world clinical settings. Moreover, body surface potential mapping leverages ECG imaging to noninvasively reconstruct epicardial potentials and create personalized digital twin hearts to achieve personalized precision medicine decision support. The topics include (1) Identification of low left ventricular ejection fraction heart failure using a single-lead ECG. (2) Identification of paroxysmal atrial fibrillation from sinus rhythm 12-lead ECG. (3) Prognosis prediction of first-year mortality in post-myocardial infarction patients using 12-lead ECG and DL. (4) Developing a personalized digital twin heart with ECG imaging technique. The topics mentioned above cover applications in various scenarios, ranging from clinical hospital to personal home usage, from models applicable to the general population to specific patient groups, and from population-wide models to personalized digital twin heart models, developing various disease diagnostic support tools from the population level to the individual level.
Appears in Collections:	[Institute of Biomedical Engineering] Electronic Thesis & Dissertation

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	40	View/Open

社群 sharing

Loading...