基於模糊神經網絡進行醫療數據評估與分類

• Main Authors: 艾迪森, Ningthoujam Avichandra Singh
• Other Authors: Yo-Ping Huang
• Format: Others
• Language: en_US
• Published: 2018
• Online Access: http://ndltd.ncl.edu.tw/handle/5k53fn

博士 === 國立臺北科技大學 === 電資學院外國學生專班 === 106 === Achieving a precise medication is the key to the modern healthcare system. Uncertainty is an inherent problem of medical diagnosis due to the lack of sufficient evidence to assess a disease. An intelligent classification model can help screen medical data and improve the classification accuracy. There have been many examinations to evaluate or identify dementia. In this research, we assess various levels of clinical dementia rating (CDR) datasets using the user-friendly computerized battery system. The battery system has 13 different examinations that were subdivided into 21 examinations to collect various datasets and give real-time results of the examination. The average examination time was designed for 10 minutes. We proposed various methods to differentiate between different CDR levels. We applied analysis of variance (ANOVA) model to analyze the differences among group means and their associated procedure, such as variation among and between groups. ANOVA uses F-tests to statistically test the equality of means. The F-test is simply a ratio of two variances. To measures reliability or internal consistency among the collected different datasets we used Cronbach’s alpha measure. Cronbach’s alpha is a measure of internal consistency, that is, how closely related a set of items are as a group. It is considered to be a measure of scale reliability. Mann–Whitney U test was used to make the comparisons to demonstrate the discriminant validity of the tablet-based assessments, separate between-group comparisons were made of the paper-based and tablet-based cognitive tests. We designed four factors from different examination models and confirmatory factor analysis (CFA) was used to verify the factor structure of the collected data. Receiver operating characteristic (ROC) analysis was used to evaluate the ability of the tablet-based tests to discriminate among different levels of CDR. The area under the curve (AUC) was represented as the primary result of the ROC analyses. Spearman’s Rank correlation coefficient technique was used to summarize the strength and direction (negative or positive) of a relationship between two variables. The results showed that there is a significant difference between test items of memory and reaction speed among groups of normal control, mild cognitive impairment (MCI) and Alzheimers disease (AD). Among these three groups, there was also a difference in test item “motor control”, while no difference in “visuospatial”. Finally, fuzzy neural network (FNN) model was proposed to classify data among different CDR levels and it achieved an accuracy of 95.20%. To verify the effectiveness of the proposed FNN model we applied it to classify two different benchmark datasets, diabetic retinopathy debrecen, and vertebral column dataset and achieved an accuracy of 98.09% and 100%, respectively.