A Study On Artificial Neural Network Prediction Model For Alzheimer's Disease In The Community Based On The Detection Of Macro And Trace Elements And Neurotransmitter

Objective: Through the field investigation of people who were older than60years in some urban community of Jiangxi Province, and the detection of somemacro and trace elements and neurotransmitters, we tried to explore and establish anartificial neural network model for Alzheimer's disease (AD) screening in thecommunity.Methods: Between2008and2010, the field study was carried out in someselected communities from Nanchang, Ji'an and Yichun. With Cluster sampling,4350persons older than60years were investigated, in which214persons werediagnosed as AD. Finally,60AD patients and60non-AD (control) persons wereselected to be the BP-ANN modeling objects. We used self-designed questionnaire tocollect epidemiological information for all checked objects, such as AD prevalenceand influencing factors, also used the Activity of Daily Living Scale (ADL) toevaluate their activities of daily living. The confirmed diagnosis of AD patients:According to the screening score of the Mini-Mental State Examination (MMSE)and combined the clinical history and signs and other related information, withreference to the International Classification of Diseases,10th edition (ICD-10) andthe Chinese Classification of Mental Disorders (CCMD-3) diagnostic criteria, thesuspected cases were confirmed finally. Then we using Clinical Dementia Rating(CDR) to evaluate the severity of AD, and using Hamminsk's ischemia rating scaleto exclude vascular dementia (VD) and mixed dementia (MID). The choice ofnon-AD: exclusion of cardiovascular disease and neurological diseases, nocongenital dementia; living in the same urban community; comparable with ADpatients; and themselves and their families willing to cooperate. In this study,10mLof venous blood was collected from each of the120modeling objects, and thecontents of Fe, Cu, Zn, AL, Ca, Se, Mn, Cr, Cd in the blood were detected by AtomicAbsorption Spectrometry, while the related neurotransmitter such as Serotonin(5-HT), Dopamine (DA) and Anti-acetylcholine receptor antibody (AchR-Ab) weredetected by radioimmunoassay (RIA). After all data collection and verification of correct, adopted SPSS13.0software to establish the database, we used theClementine12.0software to filtering the input variables, and building the ANN byrepeated simulations. Model training parameters for the ANN were: impulseitems=0.9, the initial learning rate=0.02, the maximum weight to adjust the rate ofchange is not greater than0.005; activation function is Sigmoid function. We usedareas under receiver-operating characteristic (ROC) curves (AUC), sensitivity,peculiarity and accuracy to evaluate the ANN modelfor AD prediction.Results:(1)The AD prevalence of the people who older than60years was4.92%in the community of Jiangxi province, and the female prevalence was higherthan the male. At the same time, the AD prevalence was increased with the age grow,and it approximately grows one time every five years.(2)The level of Al in the blood of AD patients was higher than the non-AD,but the level of Cr was lower than the later. While the levels of5-HT, DA andAchR-Ab in the blood of AD patients were all lower than the non-AD.(3)The architecture of BP-ANN we finally constructed was three-tier network,and the input layer was one and six nodes, which were ADL score, Cr,5-HT, age,DA and Al respectively; The hidden layer was also one and three nodes; The outputlayer was one and one node, which was AD sick.(4)The ANN which the input variables were ADL score, Cr,5-HT, age, DAand Al obtained a good effect in AD prediction, the AUC was0.929(95%CI0.868~0.968), sensitivity90.00%, specificity95.00%,92.50%of accuracy.Conclusion:(1)The AD prevalence rate was in the medium to high level inthe city community of Jiangxi Province. Compared with the Male, the Female hadmore chance to be AD patients. And the older, the more susceptible to AD.(2)TheAl element content is high in theAD patient blood, but the Cr elementcontent is low. The5-HT, DA and AchR-Ab content in the blood of AD patients wereall lower than the non-AD.(3)The ANN, with ADL score, Cr,5-HT, age, DA and Al such six variablesas input layer, has a high accuracy in Alzheimer's disease screening (prediction) inthe community. The model was simple, objective, and suitable for large-scalepopulation screening, which has a wider application prospect.