تشخیص اختلال خفیف شناختی با احداث شبکه های بیزی با داده های گمشده
|کد مقاله||سال انتشار||مقاله انگلیسی||ترجمه فارسی||تعداد کلمات|
|29089||2011||8 صفحه PDF||سفارش دهید||6697 کلمه|
Publisher : Elsevier - Science Direct (الزویر - ساینس دایرکت)
Journal : Expert Systems with Applications, Volume 38, Issue 1, January 2011, Pages 442–449
Mild Cognitive Impairment (MCI) is thought to be the prodromal phase to Alzheimer’s disease (AD), which is the most common form of dementia and leads to irreversible neurogenerative damage of the brain. In order to further improve the diagnostic quality of the MCI, we developed a MCI expert system to address MCI’s prediction and inference question, consequently, assist the diagnosis of doctor. In this system, we mainly deal with following problems: (1) Estimate missing data in the experiment by utilizing mutual information and Newton interpolation. (2) Make certain the prior feature ordering in constructing Bayesian network. (3) Construct the Bayesian network (We term the algorithm as MNBN). The experimental results indicate that MNBN algorithm achieved better results than some existing methods in most instances. The mean square error comes to 0.0173 in the MCI experiment. Our results shed light on the potential application in MCI diagnosis.
Alzheimer’s disease (AD) is the most common form of dementia and that may lead to irreversible neurogenerative damage of the brain. But the current diagnostic tools have poor sensitivity, especially for the early stages of AD and are not easy to be diagnosed until AD has led to irreversible brain damage (Morris et al., 2001). Therefore, it is very important research topic for how to diagnose AD as early as possible. Through research effort of recent 10 years, it is concluded that MCI (Mild cognitive impairment) is the early stage of the Alzheimer Diseases (Celsis, 2000, Morris et al., 2001 and Petersen et al., 2001). 10–30% of MCI patients convert to AD annually, whereas the rate of conversion of cognitively normal elderly people is 1–2% (Celsis, 2000). Furthermore, there is evidence that 100% of patients with MCI progress to greater dementia severity (Petersen et al., 2001). So the problem of diagnosing AD can be converted into the diagnosis of the MCI. Up to now, however, there is still not a strict and unified standard. In this study, we develop a specific diagnostic system on the MCI, which predicts and diagnoses the MCI by using some artificial intelligent methods. Since a practical database usually might be not complete, at first, we utilize the mutual information and Newton interpolation to estimate the values of missing data. Then, we propose to determine the feature ordering by using the mutual information and defining a “higher filter”. Finally, we construct the Bayesian network for assisting the prediction and diagnosis of the MCI. The remainder of this paper is organized as follows: Section 2 briefly reviews some related works. In Section 3, we present the MNBN algorithm. In Section 4, we further describe how to implement the MNBN algorithm. In Section 5, we report and analyze experimental results. Finally, in Section 6, we draw the main conclusions and give some discussions.
نتیجه گیری انگلیسی
In this paper, we use a set of behavioral experimental data constructing the MCI network model. We find ANT, STM and education degree are the mainly influencing factors of MCI, and get the nonlinear association among these influencing factors. The MCI system has been tested and applied in Xinhua hospital of Dalian, and we have confidence that it has potential to apply in society and ordinary family. Because each person can detect the possibility of MCI risks at home, which will greatly improve the discovery rate of MCI. If the subject is detected the high MCI risk, the system may advice the subject to go to hospital and do the further examination under the guidance of doctor. So our system might be a primary free diagnosis to MCI. In the process of constructing MCI network, we propose the MNBN algorithm, which first uses the mutual information between features to find the similar cases with the missing data, adopt Newton interpolation to estimate the missing data. Next, we utilize again the mutual information and define “higher filter” to get the suitable feature ordering. Finally, we apply the feature ordering to construct the Bayesian network. The experimental results indicate that MNBN algorithm achieves better results than other methods in most conditions. In the future, we plan to further improve the efficiency of the algorithm by following several aspects: (1) In the process of finding the similar cases, considering the complexity, we currently only compute the mutual information between two features and ignore the effects of the selected feature subset. In some conditions, if necessary, we could utilize the multivariate mutual information to further improve the accuracy. By using multivariate mutual information, we will get the sets of dependent relationships among a set of features. But it will increase the complexity of the algorithm. Therefore, how to utilize some optimal methods to increase the efficiency of the algorithm is a challenge and require further investigation intensely. (2) Using Newton interpolation to estimate the missing data demands the feature value different from each other. In order to decrease the complexity, we only compute the mean of corresponding feature values, which will make effect on the precision of the algorithm. In future, we will develop more optimal algorithm to estimate the missing data. (3) Although we gain the better performance on the above data sets, more studies will be required on how to tune the parameters, such as σ, “higher filter” and “lower filter”, which is a problem worth investigating in future research.