資訊管理學報
龔千芬;郝沛毅;
頁: 239-276
日期: 2024/04
摘要: 敗血症是一種可能導致死亡的嚴重疾病，根據世界衛生組織的統計，每年大 約有 600 萬人因為敗血症休克而死亡，死亡率高達 50％。敗血症發作的早期預 警和及早的干預治療，可以避免絕大多數的敗血症休克死亡的發生。人工智慧演 算法的蓬勃發展與重症監護病房的大量病患臨床數據的蒐集，讓應用人工智慧技 術開發早期敗血症預警系統，成為一個避免敗血症死亡的有前途的解決方案。 支持向量機可以建立最佳的分類器，深度學習可以自動學習關鍵的特徵，本 論文則是結合兩者的長處。本論文開發一種新穎的敗血症早期預警系統，它是由 全連接神經網路、雙向長短期記憶體、卷積神經網路、注意力機制、合作網路、 生成對抗網路與深層模糊支持向量機所組成。本論文使用 PhysioNet/Computing in Cardiology Challenge 2019 敗血症資料集開發一個嶄新的自調節生成對抗深度 混合神經網路架構來自動擷取關鍵的特徵，其中，卷積神經網路可以擷取與位置 無關的局部特徵，雙向長短期記憶體可以使用前向與後向運算來補抓動態訊息的 時間依賴性特徵，全連接神經網路適合分析靜態訊息內的關鍵特徵，注意力機制 可以更關注於能提升預測效能的重要特徵。此外，由合作網路與生成對抗網路所 組成的自調節雙通道模組可以生成隨機雜訊，並且增強特徵的強健性和泛化性。 最後，自調節生成對抗模型擷取的關鍵特徵，將交給新穎的模糊支持向量機來打 造最佳的敗血症預警系統，另外，我們使用深層多核心學習演算法來將模糊支持 向量機擴充到深層網路結構，以進一步提升支持向量機的推理能力。本論文開發 的預警系統可以在 6 小時之前預測敗血症的發作，並為避免敗血症休克發作導致 死亡的風險，提供理想的早期預警的解決方案。
關鍵字: 敗血症早期預測;深度學習;模糊支持向量機;醫學訊息學;生成對抗網路;
Abstract: Sepsis is a serious disease that can lead to death. According to the statistics of the World Health Organization, about 6 million people die each year due to septic shock, and the mortality rate is as high as 50%. Early warning and early intervention of sepsis onset can avoid the vast majority of septic shock deaths. The vigorous development of artificial intelligence algorithms and the collection of clinical data from a large number of patients in intensive care units have made the application of artificial intelligence technology to develop an early warning system for sepsis a promising solution to avoid sepsis deaths. Support vector machine can establish the best classifier. Deep learning algorithms can automatically learn key features. This research integrates the strengths of both models. This research proposes a novel sepsis early warning system, which is comprised of convolutional neural network, bidirectional long-term short-term memory, attention mechanism, fully connected neural network, generative adversarial network and deep fuzzy support vector machine. This research employs the PhysioNet/Computing in Cardiology Challenge 2019 sepsis dataset and develop a new self-regulating generative adversarial deep hybrid neural network architecture to automatically extract key features. Convolutional neural networks can capture positioninvariant local features. Bidirectional long-term short-term memory can use forward and backward operations to capture time-dependent features of dynamic information. Fully connected neural networks are suitable for analyzing static information. The attention mechanism can pay more attention to the important features that can improve the prediction performance. In addition, a self-regulated dual-channel module composed of a cooperative network and a generative adversarial network can generate random noise and enhance the robustness and generalization of features. Finally, the key features extracted by the self-adjusting generative adversarial model will be handed over to the novel fuzzy support vector machine to create the best sepsis early warning system. The early warning system developed in this paper can predict the onset of sepsis 6 hours earlier, and provide an ideal early warning solution to avoid the risk of death caused by septic shock
Keywords: : Sepsis early warning model, Deep learning, Fuzzy support vector machines, Medical informatics, Generative adversarial network;
瀏覽次數: 1718 下載次數: 90
引用 導入Endnote
頁: 239-276
日期: 2024/04
摘要: 敗血症是一種可能導致死亡的嚴重疾病，根據世界衛生組織的統計，每年大 約有 600 萬人因為敗血症休克而死亡，死亡率高達 50％。敗血症發作的早期預 警和及早的干預治療，可以避免絕大多數的敗血症休克死亡的發生。人工智慧演 算法的蓬勃發展與重症監護病房的大量病患臨床數據的蒐集，讓應用人工智慧技 術開發早期敗血症預警系統，成為一個避免敗血症死亡的有前途的解決方案。 支持向量機可以建立最佳的分類器，深度學習可以自動學習關鍵的特徵，本 論文則是結合兩者的長處。本論文開發一種新穎的敗血症早期預警系統，它是由 全連接神經網路、雙向長短期記憶體、卷積神經網路、注意力機制、合作網路、 生成對抗網路與深層模糊支持向量機所組成。本論文使用 PhysioNet/Computing in Cardiology Challenge 2019 敗血症資料集開發一個嶄新的自調節生成對抗深度 混合神經網路架構來自動擷取關鍵的特徵，其中，卷積神經網路可以擷取與位置 無關的局部特徵，雙向長短期記憶體可以使用前向與後向運算來補抓動態訊息的 時間依賴性特徵，全連接神經網路適合分析靜態訊息內的關鍵特徵，注意力機制 可以更關注於能提升預測效能的重要特徵。此外，由合作網路與生成對抗網路所 組成的自調節雙通道模組可以生成隨機雜訊，並且增強特徵的強健性和泛化性。 最後，自調節生成對抗模型擷取的關鍵特徵，將交給新穎的模糊支持向量機來打 造最佳的敗血症預警系統，另外，我們使用深層多核心學習演算法來將模糊支持 向量機擴充到深層網路結構，以進一步提升支持向量機的推理能力。本論文開發 的預警系統可以在 6 小時之前預測敗血症的發作，並為避免敗血症休克發作導致 死亡的風險，提供理想的早期預警的解決方案。
關鍵字: 敗血症早期預測;深度學習;模糊支持向量機;醫學訊息學;生成對抗網路;
Early detection of sepsis utilizing self-regulated generative adversarial network and deep fuzzy support vector machine
Abstract: Sepsis is a serious disease that can lead to death. According to the statistics of the World Health Organization, about 6 million people die each year due to septic shock, and the mortality rate is as high as 50%. Early warning and early intervention of sepsis onset can avoid the vast majority of septic shock deaths. The vigorous development of artificial intelligence algorithms and the collection of clinical data from a large number of patients in intensive care units have made the application of artificial intelligence technology to develop an early warning system for sepsis a promising solution to avoid sepsis deaths. Support vector machine can establish the best classifier. Deep learning algorithms can automatically learn key features. This research integrates the strengths of both models. This research proposes a novel sepsis early warning system, which is comprised of convolutional neural network, bidirectional long-term short-term memory, attention mechanism, fully connected neural network, generative adversarial network and deep fuzzy support vector machine. This research employs the PhysioNet/Computing in Cardiology Challenge 2019 sepsis dataset and develop a new self-regulating generative adversarial deep hybrid neural network architecture to automatically extract key features. Convolutional neural networks can capture positioninvariant local features. Bidirectional long-term short-term memory can use forward and backward operations to capture time-dependent features of dynamic information. Fully connected neural networks are suitable for analyzing static information. The attention mechanism can pay more attention to the important features that can improve the prediction performance. In addition, a self-regulated dual-channel module composed of a cooperative network and a generative adversarial network can generate random noise and enhance the robustness and generalization of features. Finally, the key features extracted by the self-adjusting generative adversarial model will be handed over to the novel fuzzy support vector machine to create the best sepsis early warning system. The early warning system developed in this paper can predict the onset of sepsis 6 hours earlier, and provide an ideal early warning solution to avoid the risk of death caused by septic shock
Keywords: : Sepsis early warning model, Deep learning, Fuzzy support vector machines, Medical informatics, Generative adversarial network;
瀏覽次數: 1718 下載次數: 90
引用 導入Endnote