A Study of Applying Artificial Intelligence to the Assessment of Pneumonia in Patients with Sleep Apnea

• Main Authors: LI, CHENG-TUNG, 李政東
• Other Authors: CHANG, CHUN-LANG
• Format: Others
• Language: zh-TW
• Published: 2018
• Online Access: http://ndltd.ncl.edu.tw/handle/wah7vt

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 === 106 === In the modern society, there are almost 450 million people have pneumonia and this accounts for 7 percent of the world's population for every year. In 20th century, the inventions of antibodies and vaccines improve the survivability, while lessd-eveloped countries and the patients of chronic disease, pneumonia is still high,and furthermore, it’s even the main causes of death.According to reports, it usually keeps obstructive about 20 to 40 seconds when the patients with sleep apnea was sleeping,and besides, the risk for the period when patients sleeps that should be much more higher caused by immune system infection so that fungicide is susceptible to invade for patients, also it increases the risk of pneumonia, the study shows that sleep apnea patients with higher incidence of pneumonia. This study, with the help of medical institutions to provide the patients of sleep apnea as the subjects, by the literature discussion and analysis, screening the relevant important factors, we use the Genetic Algorithm Logistic Regression, Cross-Entropy Algorithm, Particle Swarm Optimization (PSO) Algorithm and combined with Support Vector Machine (SVM), Back Propagation Neural Network, Case-based Reasoning technology, A Case-based Reasoning evaluation system and predictive models were constructed to apply evaluation to the subsequent risk with the pneumonia. The study showed that sleep apnea was a main factor for pneumonia, the top three of the related factors were Chronic Obstructive Pulmonary Disease (COPD), Tuberculosis (TB) and age, factors the higher for the weight of the factors associated with sleep apnea, the higher risk for having the pneumonia.The accuracy of the 6 predictive models in this study is over 87% and the area of the ROC curve is over 0.85. That means the models have the distinguishing ability, the models through the Friedman verification showed that all have the significant differences, and then by the T test showed that the Particle Swarm Optimization (PSO) algorithm combined with the Support Vector Machine has the best ACC and the AURC, they are 90.7% and 0.88 respectively. Three evaluation models were established in CBR Assessment system, and it was found that three models were significantly differences by Friedman verification and T test, among which the average ACC and AURC of PSO combined with Case-Based Reasoning Technology are 88.68% and 0.8564 respectively, and it’s higher than the average accuracy of GALR and CE, it is known that PSO is superior to GA and CE, these two prediction models. In this study, a test case verification system can provide medical institutions, patients and clinical workers as a reference to assist in diagnosis and evaluation.