摘要
闌尾炎是腹部疾病最常見的急性手術之一，早期病症與腹膜炎、女性生殖器疾病、消化道疾病和泌尿方面疾病的病症有極大的相似性，根據文獻的統計，依目前的臨床醫療水準仍會有25%~50%的陰性闌尾切除發生，因此，本研究的目的乃是要建立一個輔助臨床診斷之系統，就醫院觀點而言，能減少病患接受不必要的手術所衍生出來的醫療成本與醫療糾紛；醫師而言能夠降低闌尾破裂比率與陰性切除率間取得一個最佳平衡之參考依據；病患及家屬而言安撫家屬心理上的不安及降低對於手術的不確定性。
故本研究藉由使用葉振山等(2005)之個案合作機構急診科室病患資料庫為對象，建構一預測急性闌尾炎之模型，以倒傳遞類神經網路及決策樹分別建構預測急性闌尾炎之模型，再以決策樹特徵萃取出影響判別急性闌尾炎之重要因子與萃取判別之規則，將其設定為輸入變數結合倒傳遞類神經網路預測性闌尾炎之模型，其結果為決策樹結合倒傳遞類神經網路分類準確率為94.70%，ROC為0.947為最佳，其結果經過評估後可供醫師做為輔助診斷系統，找出影響判別急性闌尾炎關鍵因素進而有效降低誤診率及合併症的發生，以利提昇醫療的服務品質與成本效益。
關鍵字：急性闌尾炎、人工智慧、決策樹、倒傳遞類神經網路、特徵選取。
ABSTRACT
Appendicitis is the most common cause of acute surgical abdomen in Emergency Room (ER).In the early, the acute appendicitis are similar with peritonitis, genital diseases of female, gastrointestinal disease and urinary disease. According to the literature review the percentage of negative appendectomies has been reported to vary in 25% to50 %, thus, in this study the computer aided clinical evaluation system is expected to help the hospital to decrease the numbers unnecessary operation of the derivative cost and dissension, assist the doctor to balance the rate of ruptured or negative appendectomies and the uncertainty of operation, and reduce the anxiety for the patient and their family.
In this study, we collected 188 appendectomy cases form a comprehensive medical center located in southern Taiwan from Jehn-Shan Yeh in 2005. The important factors that influence the acute appendicitis were applied based on the index of symptoms, signs and laboratory data to establish the prediction model with Back Propagation Neural network and Decision Tree. In addition, Back Propagation Neural network combined with Decision Tree were used to establish the prediction model for acute appendicitis. The result indicates that the proposed Back Propagation Neural network combined with Decision Tree is the better method to predict acute appendicitis, and the accuracy reaches 94.70% and the ROC curve can reach 0.947.
This study provides medical professionals auxiliary reference for evaluating acute appendicitis of clinical diagnosis and finds the important factors that influence the acute appendicitis. Meanwhile, the rate of delayed or inaccurate diagnosis of acute appendicitis is reduced and the substantial enhancement of medical service quality and cost efficiency are acquired.
Keywords: Acute Appendicitis; Back Propagation Neural Network; Decision Tree; Feature Selection; Artificial Intelligence
|