A Study of Lung Cancer in Patients with Diabetes

• Main Authors: LIU, YU-SHENG, 劉宇晟
• Other Authors: CHANG, CHUN-LANG
• Format: Others
• Language: zh-TW
• Published: 2018
• Online Access: http://ndltd.ncl.edu.tw/handle/6bk626

碩士 === 國立虎尾科技大學 === 工業管理系工業工程與管理碩士班 === 107 === In recent years, both Taiwan’s epidemiological and global statistics show an increasing trend in the incidence and prevalence of diabetes and cancer, as diabetes and lung cancer are both chronic diseases. According to the statistics by the Ministry of Health and Welfare in 2017, malignant tumor (cancer) and diabetes are respectively ranked the first and the fifth among the top ten causes of mortality, in which lung cancer is the leading cause of cancer death. Based on the epidemiological report, diabetes may be the high-risk factor of poor prognosis and the cause of tumor incidence. This study used medical records of diabetic patients from the database of a medical institution in Taiwan in recent years as the primary research subjects. Through literature collection and interviews with professional physicians, risk factors, which affect diabetic patients from getting lung cancer were selected. The weight of each factor was calculated via Genetic Algorithms-Logistic Regression (GALR), Cross Entropy (CE) and Particle Swarm Optimization (PSO). The weighted values were then combined with Back Propagation Neural Network (BPN), Support Vector Machines (SVM) and Case-Based Reasoning (CBR) to establish a prediction and evaluation model. Lastly, through cross-validation and statistical analysis, differences among models were assessed. The average accuracy (ACC) of the six prediction models in this study was above 89% and area under ROC curve (AURC) was also above 0.83. There was a significant difference in Friedman test, where PSO+SVM showed the best result. After the t-test, the three prediction models, PSO+BPN, PSO+SVM and GALR+SVM, showed no significant difference, suggesting an equal predictability of the models. In terms of evaluation, these systems showed significant difference in Friedman test. The t-test showed statistical difference in these three algorithms with CBR. PSO+CBR was the best evaluation system with an average ACC at 91.44% and an average AURC at 0.882. This study could provide as reference for medical evaluation by medical institutions, patients and clinicians.