臺灣博碩士論文加值系統

本研究探討使用先進的資料探勘技術發覺術後噁心和嘔吐 (Postoperative Nausea and Vomiting, PONV) 的影響因子，應用機器、深度學習的演算法，確定了導致PONV的關鍵因素，不只通過嚴格的資料前處理和特徵選擇的方式，還提出新的方法，能夠在較少專家知識輔助的情況下，使用ChatGPT協助分析非結構化的文本，本文還比較了幾種不同的文本處理技術會如何影響模型的效能，結果發現，在使用LightGBM結合本文提出的Score system效果最好，在模型的解釋方面，重要因子排序為，患者的性別、手術的麻醉時間、術後止痛藥嗎啡劑量、是否全靜脈麻醉特徵，Score system提供了更細緻的PONV風險理解，在所有手術類型中，下顎上升枝垂直截骨術、改良式乳房根除手術、部份乳房切除術PONV的發生率最高，通過這些步驟，進一步找出不同的手術科別中，較容易與較不容易發生PONV的手術類型，為臨床設置中的預測建模和預防策略奠定了基礎。

This research investigates the determinants of Postoperative Nausea and Vomiting (PONV) using advanced data mining techniques, including machine and deep learning algorithms. We identified critical factors contributing to PONV through rigorous data preprocessing and feature selection, introducing a novel method that requires minimal expert knowledge. Notably, the study utilizes ChatGPT to analyze unstructured text data, evaluating various text processing techniques and their impact on model performance. The results indicate that the combination of LightGBM and our proposed Score system yields the most effective outcomes. Key factors influencing PONV include patient gender, anesthesia duration, postoperative morphine dosage, and the use of total intravenous anesthesia. The Score system enhances our understanding of PONV risks by detailing specific surgical procedures that exhibit high incidence rates, such as vertical ramus osteotomy, modified radical mastectomy, and partial mastectomy. Our findings establish a foundation for predictive modeling and preventive strategies in clinical settings, identifying surgeries within different specialties that are more or less likely to result in PONV.

目錄
論文審定書 i
摘要 ii
Abstract iii
目錄 iv
圖目錄 vii
表目錄 viii
第一章 緒論 1
4.1 研究背景 1
4.2 研究問題 1
4.3 研究目標 2
第二章 相關方法與研究 4
5.1 噁心與嘔吐的判斷 4
5.2 術後噁心與嘔吐的因子 4
5.2.1 麻醉關聯因子 4
5.2.2 病患關聯因子 5
5.2.3 手術關聯因子 5
5.3 PONV預測模型 6
5.4 分類演算法介紹 8
5.4.1 邏輯迴歸演算法(Logistic regression) 8
5.4.2 隨機森林(Random Forest) 8
5.4.3 極限梯度提升 (eXtreme Gradient Boosting, XGBoost) 9
5.4.4 輕量梯度提升 (Light Gradient Boosting Machine, LightGBM) 10
5.4.5 類別梯度提升 (Categorical boosting, CatBoost) 10
5.4.6 TabTrasformer 11
5.4.7 FT-Transformer 12
5.4.8 TabNet 13
第三章 研究方法 15
6.1 藥物資料處理 16
6.2 資料前處理 17
6.2.1 不平衡資料 17
6.2.2 異常值處理 18
6.2.3 資料洩漏 19
6.2.4 資料正規化 20
6.2.5 特徵選擇 20
6.2.6 混淆效應 24
6.2.7 遞迴特徵消除 25
6.2.8 模型的特徵重要性 26
6.3 非結構化文本資料處理技術的比較 27
6.3.1 Ordinal encoding 27
6.3.2 Scoring system 27
6.3.3 LangChain QA system 29
6.4 模型的選擇建立 30
6.4.1 機器學習模型 30
6.4.2 深度學習模型 31
6.5 模型可解釋性 32
6.5.1 基於Scoring System的關鍵詞解釋 32
6.5.2 SHAP值的模型重要性與依賴圖 32
6.5.3 LangChain解釋 33
第四章 實驗結果與討論 34
7.1 資料收集 34
7.2 實驗設計與評量準則 34
7.3 PONV 預測之實驗說明 36
7.3.1 實驗之屬性 36
7.3.2 Scoring System實驗說明 36
7.3.3 LangChain QA System實驗說明 37
7.4 測試集的實驗結果 39
7.4.1 資料處理方法的比較 42
7.4.2 模型的性能的比較 43
7.5 模型的解釋與討論 44
7.5.1 Scoring System的關鍵詞解釋 44
7.5.2 SHAP值分析 52
7.5.3 LangChain QA System解釋 54
7.5.4 乳房外科與骨科手術的PONV分析 56
第五章 研究貢獻與未來展望 60
8.1 總結 60
8.2 研究貢獻 60
8.3 未來展望 61
參考文獻 62
附錄 64

[1]T. J. Gan, "Risk factors for postoperative nausea and vomiting," Anesthesia & Analgesia, vol. 102, no. 6, pp. 1884-1898, 2006.
[2]J. Lerman, "Surgical and patient factors involved in postoperative nausea and vomiting," British Journal of Anaesthesia, vol. 69, pp. 24S-24S, 1992.
[3]C. Apfel et al., "Evidence-based analysis of risk factors for postoperative nausea and vomiting," British journal of anaesthesia, vol. 109, no. 5, pp. 742-753, 2012.
[4]R. Sukhani, J. Vazquez, A. L. Pappas, K. Frey, M. Aasen, and S. Slogoff, "Recovery after propofol with and without intraoperative fentanyl in patients undergoing ambulatory gynecologic laparoscopy," Anesthesia & Analgesia, vol. 83, no. 5, pp. 975-981, 1996.
[5]J. Rose and M. Watcha, "Postoperative nausea and vomiting in paediatric patients," British journal of anaesthesia, vol. 83, no. 1, pp. 104-117, 1999.
[6]P. Kranke et al., "An increased body mass index is no risk factor for postoperative nausea and vomiting: A systematic review and results of original data Note," Acta Anaesthesiologica Scandinavica, vol. 45, no. 2, pp. 160-166, 2001.
[7]C. Apfel et al., "Volatile anaesthetics may be the main cause of early but not delayed postoperative vomiting: a randomized controlled trial of factorial design," British journal of anaesthesia, vol. 88, no. 5, pp. 659-668, 2002.
[8]C. C. Apfel et al., "A factorial trial of six interventions for the prevention of postoperative nausea and vomiting," New England journal of medicine, vol. 350, no. 24, pp. 2441-2451, 2004.
[9]D. R. Sinclair, F. Chung, and G. Mezei, "Can postoperative nausea and vomiting be predicted?," The Journal of the American Society of Anesthesiologists, vol. 91, no. 1, pp. 109-118, 1999.
[10]M. M. Cohen, P. G. Duncan, D. P. DeBoer, and W. A. Tweed, "The postoperative interview: assessing risk factors for nausea and vomiting," Anesthesia & Analgesia, vol. 78, no. 1, pp. 7-16, 1994.
[11]M. Koivuranta, E. Läärä, L. Snåre, and S. Alahuhta, "A survey of postoperative nausea and vomiting," Anaesthesia, vol. 52, no. 5, pp. 443-449, 1997.
[12]C. Apfel and N. Roewer, "Riskfactors for nausea and vomiting after general anaesthesia: fiction and facts," Der Anaesthesist, vol. 49, no. 7, pp. 629-642, 2000.
[13]M. Palazzo and R. Evans, "Logistic regression analysis of fixed patient factors for postoperative sickness: a model for risk assessment," British journal of anaesthesia, vol. 70, no. 2, pp. 135-140, 1993.
[14]C. Apfel et al., "A risk score to predict the probability of postoperative vomiting in adults," Acta Anaesthesiologica Scandinavica, vol. 42, no. 5, pp. 495-501, 1998.
[15]D. R. Cox, "Regression models and life‐tables," Journal of the Royal Statistical Society: Series B (Methodological), vol. 34, no. 2, pp. 187-202, 1972.
[16]L. Breiman, "Random forests," Machine learning, vol. 45, pp. 5-32, 2001.
[17]A. Cutler, D. R. Cutler, and J. R. Stevens, "Random forests," Ensemble machine learning: Methods and applications, pp. 157-175, 2012.
[18]G. Ke et al., "Lightgbm: A highly efficient gradient boosting decision tree," Advances in neural information processing systems, vol. 30, 2017.
[19]L. Prokhorenkova, G. Gusev, A. Vorobev, A. V. Dorogush, and A. Gulin, "CatBoost: unbiased boosting with categorical features," Advances in neural information processing systems, vol. 31, 2018.
[20]X. Huang, A. Khetan, M. Cvitkovic, and Z. Karnin, "Tabtransformer: Tabular data modeling using contextual embeddings," arXiv preprint arXiv:2012.06678, 2020.
[21]Y. Gorishniy, I. Rubachev, V. Khrulkov, and A. Babenko, "Revisiting deep learning models for tabular data," Advances in Neural Information Processing Systems, vol. 34, pp. 18932-18943, 2021.
[22]S. Ö. Arik and T. Pfister, "Tabnet: Attentive interpretable tabular learning," in Proceedings of the AAAI conference on artificial intelligence, 2021, vol. 35, no. 8, pp. 6679-6687.
[23]S. Zhang, "Nearest neighbor selection for iteratively kNN imputation," Journal of Systems and Software, vol. 85, no. 11, pp. 2541-2552, 2012.
[24]Q. Chen, Z. Meng, X. Liu, Q. Jin, and R. Su, "Decision variants for the automatic determination of optimal feature subset in RF-RFE," Genes, vol. 9, no. 6, p. 301, 2018.
[25]U. Ruby and V. Yendapalli, "Binary cross entropy with deep learning technique for image classification," Int. J. Adv. Trends Comput. Sci. Eng, vol. 9, no. 10, 2020.