- 學位論文
使用卷積神經網絡運用在急診12導程心電圖上預測急性心肌梗塞後的死亡率
Convolutional neural network prediction of mortality after acute myocardial infarction by using the emergency department 12-lead electrocardiogram
本文將於2026/12/11開放下載。若您希望在開放下載時收到通知，可將文章加入收藏。
摘要
在急性心肌梗塞（AMI）分級過程中使用 12導程心電圖（12-lead ECG），可準確預測長期影響，為全方面照護決策提供訊息，應用於現有數據的創新方法為AMI預後提供了進一步的見解。為了增強評估長期風險預後方面的價值，在先進的機器學習技術上使用具有擴展多元性的12-lead ECG，並辨別預測因素和結果之間錯綜複雜的關聯。 這項回顧性對照研究使用國立台灣大學醫院的電子病歷來識別 2009 年 1 月 1 日至 2016 年 12 月 31 日期間首次發生急性心肌梗塞的患者。根據分級時的12-lead ECG，運用卷積神經網絡（CNN）該模型開發並驗證了預測 30 天和 360 天的死亡率。此模型與傳統預測評分Thrombolysis in myocardial infarction (TIMI) and Global Registry of Acute Coronary Events (GRACE) 進行了準確性的比較。研究期間共涵蓋 4146 名患者（平均 [標準偏差] 年齡，66 [13] 歲，3081 名 (74.3%) 男性）。 在30天死亡率的預測中，GRACE評分和CNN模型相比TIMI評分高（C統計量，GRACE和CNN模型分別為0.750和0.803，TIMI為0.569）。在 360 天的長期預測結果，CNN 模型比其他兩個傳統模型預測更準確（C 統計量，CNN 模型為 0.814）。 在子群分析中，CNN 模型對 65 歲以下的男性患者顯示更好的預測能力（比值比 15.02，95% 機率樣本的信賴區間 [CI] 6.97-30.9）。那些對死亡率的預測有五倍於對於生存力的預測率（風險比，5.22；95% CI 3.38-8.06）。 此項對照研究，機器學習模型在預測 AMI 後 360 天死亡率方面取得了顯著的進步，特別是對於年輕男性患者。 將此模型應用於12-lead ECG是一項普遍性和低成本的測試，使得12-lead ECG成為預測AMI 患者預後有力的預測工具。
並列摘要
The utilization of the 12-lead electrocardiogram (ECG) during acute myocardial infarction (AMI) triage can accurately predict long-term effects, informing comprehensive care decisions. Innovative methods applied to existing data offer further insights into AMI prognosis. To assess the value of advanced machine learning techniques in enhancing long-term risk prognostication, employing a 12-lead ECG with an expanded set of variables, and discerning intricate associations between predictors and results. This retrospective cohort study used electronic medical records from National Taiwan University Hospital to identify patients with first AMI between January 1, 2009, and December 31, 2016. The deep learning (DL) model was formulated and validated to predict 30-day and 360-day mortality based on 12-lead ECG at triage. The model was constructed upon a convolutional neural network (CNN). Its accuracy in comparison to the traditional predictive scoring systems, with Thrombolysis in myocardial infarction (TIMI) and Global Registry of Acute Coronary Events (GRACE) risk scores. A total of 4146 patients (mean [SD] age, 66 [13] years, 3081 (74.3%) male) were identified during the research period. In the prediction of 30-day mortality, the GRACE score and CNN model improved predictive ability compared with the TIMI score (C statistic, 0.750 and 0.803 for GRACE and CNN model respectively vs 0.569 for TIMI). For a 360-day long-term outcome, the CNN model predicted more precisely than the other two conventional models (C statistic, 0.814 for the CNN model). The subgroup analysis showed that the CNN model provided better analytical ability in male patients below 65 years old (odd ratio 15.02, 95% confidence interval [CI] 6.97-30.9). Individuals who received a positive prognosis for mortality faced a hazard ratio of 5.22 (95% CI 3.38-8.06), indicating a fivefold higher risk of future mortality compared to those expected to survive. In this cohort study, the DL model was allied with a respectable improvement in the prediction of 360-day mortality after AMI, especially for young male patients. Utilizing this model with the 12-lead ECG, a widely available and economical test, enables the ECG to function as a potent predictive tool among AMI patients.
參考文獻
Alabas, O. A., Jernberg, T., Pujades-Rodriguez, M., Rutherford, M. J., West, R. M., Hall, M., Timmis, A., Lindahl, B., Fox, K. A. A., Hemingway, H., & Gale, C. P. (2020). Statistics on mortality following acute myocardial infarction in 842 897 Europeans. Cardiovasc Res, 116(1), 149-157. https://doi.org/10.1093/cvr/cvz197
Al-Zaiti, S. S., Martin-Gill, C., Zègre-Hemsey, J. K., Bouzid, Z., Faramand, Z., Alrawashdeh, M. O., Gregg, R. E., Helman, S., Riek, N. T., Kraevsky-Phillips, K., Clermont, G., Akcakaya, M., Sereika, S. M., Van Dam, P., Smith, S. W., Birnbaum, Y., Saba, S., Sejdic, E., & Callaway, C. W. (2023). Machine learning for ECG diagnosis and risk stratification of occlusion myocardial infarction. Nature medicine, 29(7), 1804–1813. https://doi.org/10.1038/s41591-023-02396-3
Antman, E. M., Cohen, M., Bernink, P. J., McCabe, C. H., Horacek, T., Papuchis, G., Mautner, B., Corbalan, R., Radley, D., & Braunwald, E. (2000). The TIMI risk score for unstable angina/non-ST elevation MI: A method for prognostication and therapeutic decision making. JAMA, 284(7), 835-842. https://doi.org/10.1001/jama.284.7.835
Assessment of the, S., Efficacy of a New Thrombolytic, I., Van De Werf, F., Adgey, J., Ardissino, D., Armstrong, P. W., Aylward, P., Barbash, G., Betriu, A., Binbrek, A. S., Califf, R., Diaz, R., Fanebust, R., Fox, K., Granger, C., Heikkila, J., Husted, S., Jansky, P., Langer, A., . . . White, H. (1999). Single-bolus tenecteplase compared with front-loaded alteplase in acute myocardial infarction: the ASSENT-2 double-blind randomised trial. Lancet, 354(9180), 716-722. https://doi.org/10.1016/s0140-6736(99)07403-6
Attia, Z.I., Kapa, S., Lopez-Jimenez, F. et al. Screening for cardiac contractile dysfunction using an artificial intelligence–enabled electrocardiogram. Nat Med 25, 70–74 (2019). https://doi.org/10.1038/s41591-018-0240-2