糖尿病患者罹患膀胱癌之評估研究

A Study of Bladder Cancer in Patients with Diabetes

陳昱安(Yu-An Chen)

指導教授 ： 張俊郎

國立虎尾科技大學/管理學院/工業管理系工業工程與管理碩士班/碩士(2018年)

摘要

根據衛生福利部2014年的統計，膀胱癌為國人癌症死因第9位。即使接受治療，膀胱癌復發率仍高達70%，必須確保完全根除。膀胱癌是泌尿系統中常見的惡性疾病，通常60歲以上男性較容易罹患，而男與女患者的比例約為2.7：1。根據相關研究，與糖尿病有正相關的癌症包括大腸直腸癌、肝癌、膀胱癌等，而膀胱癌的死亡率在糖尿病患相對風險更高。
本研究利用醫療機構提供的資料庫，透過相關文獻進行探討，篩選出有關的重要因子，運用交叉熵演算法、基因邏輯斯演算法、粒子群最佳化演算法結合倒傳遞類神經、支援向量機與案例式推理技術，建構預測模型與案例式推理評估系統，評估糖尿病是否有伴隨膀胱癌之風險，給予病患或所需者做為參考依據，以預測評估其風險，在發生前做有效控制，減緩發生時間及降低發生率。
本研究果顯示糖尿病為罹患膀胱癌的危險因子，權重值排名前三位分別為高血壓、年齡與尿路結石，根據糖尿病患者罹患有關因子權重值越高，會使罹患膀胱癌的風險增加。本研究之6個預測模型ACC皆達88%以上，且AURC也在0.86以上，模型經傅利曼檢定皆有顯著差異，其中以粒子群最佳化演算法結合支援向量機之ACC與AURC表現較佳分別為93.018%與0.9114。
在CBR評估模型之3個評估系統中，經傅利曼檢定，PSO與GALR並無顯著差異，但皆與CE存在顯著差異性，兩系統之ACC皆達93%以上，且AURC也在0.92以上。利用測試案例之驗證系統，可以提供醫療相關機關、患者及臨床工作者做為輔助診斷評估之參考依據。

關鍵字

糖尿病 ； 膀胱癌 ； 粒子群最佳化演算法 ； 交叉熵演算法 ； 基因邏輯斯迴歸演算法 ； 倒傳遞類神經網路 ； 支援向量機 ； 案例式推理系統

並列摘要

According to the 2014 statistics of the Ministry of Health and Welfare, bladder cancer is the ninth leading cause of cancer death. Even after treatment, the recurrence rate of bladder cancer is still as high as 70% to ensure complete eradication. Bladder cancer is a common malignant disease in the urinary system. Men are generally more likely to suffer from over the age of 60, while the ratio of men to women is about 2.7: 1. According to related research, cancers that are positively correlated with diabetes include colorectal cancer, liver cancer, bladder cancer, and the like, whereas the mortality of bladder cancer is at a relatively higher risk in diabetic patients.
In this study, using the database provided by medical institutions, we explored the relevant documents and screened out relevant important factors. By using cross-entropy algorithm, gene-logic algorithm and particle swarm optimization algorithm, Machine and case-based reasoning techniques, a predictive model and a case-based reasoning assessment system to assess whether diabetes has a risk of bladder cancer, and to give patients or those who need it as a reference for predicting and assessing their risk before they occur Effective control, slow down the occurrence time and reduce the incidence.
This study shows that diabetes mellitus is a risk factor for bladder cancer. The top three weights are hypertension, age and urinary tract stones. The higher the risk factor of diabetes-related factors, the higher the risk of bladder cancer. In this study, all of the six prediction models reached 88%, and the AURC was above 0.86. The models were all significantly different by the Fourierist test. Particle swarm optimization algorithm combined with the support vector machine ACC and AURC Preferably 93.018% and 0.9114 respectively.
Among the three assessment systems of CBR assessment model, there was no significant difference between PSO and GALR, but all were significantly different from CE. The ACC of the two systems reached more than 93%, and the AURC was above 0.92 . The use of test case validation system can provide medical-related agencies, patients and clinicians as a diagnostic reference to assist diagnosis.

並列關鍵字

Diabetes ； Bladder cancer ； Particle swarm optimization algorithm ； Cross entropy algorithm ； Genetic logistic regression algorithm ； Inverted delivery neural network ； Support vector machine ； Case-based reasoning system

參考文獻

楊士杰、李坤洲(2012)。粒子群演算法應用於線性天線陣列最佳化。華岡工程學報，第29期，頁109-114。

林文賜、廖紹安、張文詔、周建明、洪耀明 (2008)，「應用支援向量機於九份二山崩塌地變遷評估之研究」，水保技術，第3卷，第3期，頁108-116。

林明傑、董子毅 (2008)，「危險評估中ROC曲線在預測2×2表上與敏感度及特異度之關係」，亞洲家庭暴力與性侵害期刊卷，第4卷，第2期，頁64-74。

黃宇翔、王百祿 (2008)，「ARIMA 與適應性SVM之混合模型於股價指數預測之研究」，電子商務學報，第10卷，第4期，頁1041-1065。

蔡顯麞、劉湘川(2009)，「兩種脊回歸模式與複線性回歸模式之交互驗證比較」，測驗統計年刊，第17期，第77-83頁。