簡易檢索 / 詳目顯示

回結果列表
研究生： 林璟任
Lin, Ching-Jen
論文名稱： 結合大型語言模型與臨床文本分析之住院病患照護與診斷支援系統
LLM-Driven Diagnostic Support for Inpatient Care via Clinical Notes
指導教授： 鄧維光
Teng, Wei-Teng
學位類別： 碩士
Master
系所名稱： 工學院 - 工程科學系
Department of Engineering Science
論文出版年： 2025
畢業學年度： 113
語文別： 中文
論文頁數： 75
中文關鍵詞： 早期預警系統 、普通病房 、深度學習 、生命徵象 、大型語言模型 、臨床文本 、非結構化資料 、泌尿道感染 、提示工程
外文關鍵詞： early warning system, general wards, deep learning, vital signs, large language modelsnical text, unstructured data, urinary tract infection, clinical text, unstructured data, urinary tract infection, prompt engineering
相關次數： 點閱：55  下載：0 
分享至:
查詢本校圖書館目錄 查詢臺灣博碩士論文知識加值系統 勘誤回報

    • 在臨床決策過程中，準確且即時的判斷對於提升病患照護品質與降低併發風險至關重要。然而，醫療現場普遍面臨人力短缺與資訊整合困難的挑戰，使得高品質決策更加困難。特別是臨床紀錄中大量非結構化資料 (如出院病摘、醫師紀錄等)，雖然包含診斷與病史等關鍵線索，卻因格式不一與語意多樣而難以由傳統方法自動處理與理解，限制了其於臨床決策支援系統中的應用價值。為了回應此一挑戰，本研究提出一套能有效整合結構化與非結構化臨床資料的智慧決策架構，應用大型語言模型處理文本資訊，並以分類任務形式解決兩項具代表性的臨床問題。針對泌尿道感染分類，我們設計樹狀決策流程，使LLM能依照臨床邏輯分層進行推理，並結合提示設計控制語意方向與任務一致性，進一步提升模型穩定性與可解釋性。實驗顯示，此架構能有效改善分類性能，並提供具邏輯依據的診斷輔助資訊。在病患惡化預測任務中，我們將非結構化文本轉換為結構化病史特徵，並與生命徵象等資料整合至混合模型中，提升模型對潛在高風險病患的辨識能力。結果顯示，在多種時間窗口下，導入文本資訊可顯著提升敏感度與整體預測準確率，展現語言模型對臨床上下文理解的輔助價值。總結來說，本研究提出的結合資料轉換、邏輯結構與語言模型控制策略的智慧分類架構，不僅提升模型效能與臨床實用性，也展示出LLM於異質資料處理與決策輔助上的應用潛力，為未來臨床人工智慧系統的發展提供一具參考價值的實踐路徑。

    Accurate and timely decision-making is critical in clinical care but often hindered by workforce shortages and challenges in integrating heterogeneous data. A large portion of clinical information exists in unstructured formats, such as discharge summaries and physician notes, which are rich in diagnostic insights yet difficult to process with traditional methods. This work proposes a framework that integrates structured and unstructured clinical data using large language models to support classification-based clinical decision-making. For urinary tract infection classification, we design a tree-structured reasoning process guided by prompt engineering to improve model stability and interpretability. Experimental results show improved classification performance with more consistent and explainable outputs. In the deterioration prediction task, we transform clinical text into structured features and combine them with vital signs in a hybrid model. Incorporating textual information significantly improves sensitivity and prediction accuracy across time windows. Overall, our approach demonstrates that LLMs, when paired with thoughtful data transformation and reasoning strategies, can effectively handle diverse clinical data and enhance decision support in real-world settings.

    Chapter 1 Introduction 1 1.1 Motivation and Overview 1 1.2 Challenges of This Work 1 Chapter 2 Preliminaries 4 2.1 Medical Diagnosis as a Classification Problem 4 2.2 Importance of Utilizing Multiple Data Sources in Diagnosis 5 2.3 The Potential and Challenges of Large Language Models in Clinical Applications 9 2.3.1 Potential of Utilizing LLMs in Diagnosis 9 2.3.2 Applications and Challenges of LLMs in Diagnosis 10 2.3.3 Methods to Enhance LLM Performance 10 2.4 Applications and Challenges of Classification in Diagnosis 12 2.5 Integrating LLMs into an AI Pipeline for Healthcare Classification Applications 13 Chapter 3 Our Proposed Scheme 15 3.1 Design of Our Proposed Scheme 15 3.2 Classification of Urinary Tract Infection Using Large Language Models 16 3.2.1 Challenges of UTI Detection in Clinical Practice 16 3.2.2 UTI Classification via Tree-Structured Thought Process Design 17 3.2.3 Designing Prompts for Decision Points 23 3.3 Classification of Clinical Deterioration Using the Hybrid Model 25 3.3.1 Challenges in Predicting Clinical Deterioration 25 3.3.2 Integrating Unstructured Features to Enhance Clinical Classification Models 25 3.3.3 Combining Vital Signs with Binary Labels 26 3.3.4 Combining Vital Signs with Diagnosis Advice 31 Chapter 4 Experimental Results 34 4.1 Classification Results for Urinary Tract Infection 35 4.1.1 Dataset for UTI Diagnosis 35 4.1.2 Multi-class Classification Results: Comparison of Baseline and Tree-Structured Frameworks 37 4.1.3 Comparison of Performance between Multi-class and Binary classification 38 4.1.4 Comparison of LLM Classification Performance With and Without the “Uncertain” Option 40 4.1.5 LLM Cost Comparison 42 4.1.6 LLM Reasoning for UTI 43 4.2 Classification Results for Clinical Deterioration 46 4.2.1 Dataset of Clinical Deterioration Prediction 46 4.2.2 Performance of LSTM Using Vital Signs and Disease Labels 47 4.2.3 Performance of Hybrid model Using Vital Signs and Diagnosis Advice 52 4.3 Result Discussion 55 Chapter 5 Conclusions and Future Work 56 Bibliography 59

    [1] Alowais, S. A., Alghamdi, S. S., Alsuhebany, N., Alqahtani, T., Alshaya, A. I., Almohareb, S. N., ... & Albekairy, A. M. (2023). Revolutionizing healthcare: the role of artificial intelligence in clinical practice. BMC medical education, 23(1), 689.
    [2] Kaur, R., Ginige, J. A., & Obst, O. (2021). A systematic literature review of automated ICD coding and classification systems using discharge summaries. arXiv preprint arXiv:2107.10652.
    [3] Fink, M. A., Bischoff, A., Fink, C. A., Moll, M., Kroschke, J., Dulz, L., ... & Weber, T. F. (2023). Potential of ChatGPT and GPT-4 for data mining of free-text CT reports on lung cancer. Radiology, 308(3), e231362.
    [4] Kwon, J. M., Lee, Y., Lee, Y., Lee, S., & Park, J. (2018). An algorithm based on deep learning for predicting in‐hospital cardiac arrest. Journal of the American Heart Association, 7(13), e008678.
    [5] Ozkan, I. A., Koklu, M., & Sert, I. U. (2018). Diagnosis of urinary tract infection based on artificial intelligence methods. Computer methods and programs in biomedicine, 166, 51-59.
    [6] Kaieski, N., Da Costa, C. A., da Rosa Righi, R., Lora, P. S., & Eskofier, B. (2020). Application of artificial intelligence methods in vital signs analysis of hospitalized patients: A systematic literature review. Applied Soft Computing, 96, 106612.
    [7] Zhang, D., Yin, C., Zeng, J., Yuan, X., & Zhang, P. (2020). Combining structured and unstructured data for predictive models: a deep learning approach. BMC medical informatics and decision making, 20, 1-11.
    [8] Omiye, J. A., Gui, H., Rezaei, S. J., Zou, J., & Daneshjou, R. (2024). Large language models in medicine: the potentials and pitfalls: a narrative review. Annals of internal medicine, 177(2), 210-220.
    [9] Wang, J., Shi, E., Yu, S., Wu, Z., Ma, C., Dai, H., ... & Zhang, S. (2023). Prompt engineering for healthcare: Methodologies and applications. arXiv preprint arXiv:2304.14670.
    [10] Yao, S., Yu, D., Zhao, J., Shafran, I., Griffiths, T., Cao, Y., & Narasimhan, K. (2023). Tree of thoughts: Deliberate problem solving with large language models. Advances in neural information processing systems, 36, 11809-11822.
    [11] Taylor, R. A., Moore, C. L., Cheung, K. H., & Brandt, C. (2018). Predicting urinary tract infections in the emergency department with machine learning. PloS one, 13(3), e0194085.
    [12] Jeng, S. L., Huang, Z. J., Yang, D. C., Teng, C. H., & Wang, M. C. (2022). Machine learning to predict the development of recurrent urinary tract infection related to single uropathogen, Escherichia coli. Scientific reports, 12(1), 17216.
    [13] Cho, K. J., Kwon, O., Kwon, J. M., Lee, Y., Park, H., Jeon, K. H., ... & Oh, B. H. (2020). Detecting patient deterioration using artificial intelligence in a rapid response system. Critical care medicine, 48(4), e285-e289.
    [14] Fonseca, J., Liu, X., Oliveira, H. P., & Pereira, T. (2023). Mortality prediction using medical time series on TBI patients. Computer methods and programs in biomedicine, 242, 107806.
    [15] De Angeli, K., Gao, S., Danciu, I., Durbin, E. B., Wu, X. C., Stroup, A., ... & Yoon, H. J. (2022). Class imbalance in out-of-distribution datasets: Improving the robustness of the TextCNN for the classification of rare cancer types. Journal of biomedical informatics, 125, 103957.
    [16] Pornprasit, C., & Tantithamthavorn, C. (2024). Fine-tuning and prompt engineering for large language models-based code review automation. Information and Software Technology, 175, 107523.
    [17] Timm, M. R., Russell, S. K., & Hultgren, S. J. (2025). Urinary tract infections: pathogenesis, host susceptibility and emerging therapeutics. Nature Reviews Microbiology, 23(2), 72-86.
    [18] Møller, J. K., Sørensen, M., & Hardahl, C. (2021). Prediction of risk of acquiring urinary tract infection during hospital stay based on machine-learning: A retrospective cohort study. PloS one, 16(3), e0248636.
    [19] Iscoe, M., Socrates, V., Gilson, A., Chi, L., Li, H., Huang, T., ... & Taylor, R. A. (2024). Identifying signs and symptoms of urinary tract infection from emergency department clinical notes using large language models. Academic Emergency Medicine, 31(6), 599-610.
    [20] Geerligs, L., Rankin, N. M., Shepherd, H. L., & Butow, P. (2018). Hospital-based interventions: a systematic review of staff-reported barriers and facilitators to implementation processes. Implementation Science, 13, 1-17.
    [21] Goh, K. H., Wang, L., Yeow, A. Y. K., Poh, H., Li, K., Yeow, J. J. L., & Tan, G. Y. H. (2021). Artificial intelligence in sepsis early prediction and diagnosis using unstructured data in healthcare. Nature communications, 12(1), 711.
    [22] Gao, Y., Miller, T., Xu, D., Dligach, D., Churpek, M. M., & Afshar, M. (2022, October). Summarizing patients’ problems from hospital progress notes using pre-trained sequence-to-sequence models. In Proceedings of COLING. International Conference on Computational Linguistics (Vol. 2022, p. 2979).
    [23] Zhang, X., Talukdar, N., Vemulapalli, S., Ahn, S., Wang, J., Meng, H., ... & Chen, B. (2024). Comparison of prompt engineering and fine-tuning strategies in large language models in the classification of clinical notes. AMIA Summits on Translational Science Proceedings, 2024, 478.
    [24] Møller, J. K., Sørensen, M., & Hardahl, C. (2021). Prediction of risk of acquiring urinary tract infection during hospital stay based on machine-learning: A retrospective cohort study. PloS one, 16(3), e0248636.

    無法下載圖示 校內：2030-08-29公開
    校外：2030-08-29公開
    電子論文尚未授權公開，紙本請查館藏目錄
    QR CODE