The Characteristics Of Early Warning For Deteriorative MCI Of The Aged In Chuansha Community Based On Machine Learning

Mild Cognitive Impairment(MCI),an early form of Alzheimer’s Disease(AD),can have approximately one-third of the cases deteriorate to dementia.There are limited treatments for dementia.Therefore,it is of great significance to screen and follow up the MCI population,establish a prediction model for the deteriorative MCI,and determine its early warning characteristics for early clinical detection for early intervention.Previous studies have shown that Amyloid β protein(A β)and Phosphorylated tau(P-tau)in cerebrospinal fluid and PET brain imaging are the important biological diagnostic markers for AD,and the Apolipoprotein E(APOE)4 gene plays a certain role in the development of AD,but its prediction effect for rapidly progressive MCI is not certified,and additionally,the approach and cost made it’s implementations difficult.Brain magnetic resonance imaging(MRI)and serum Aβ and APOE gene detection,characterized by easy implementation in clinical practice,are expected to be potential biomarkers for predicting disease deterioration.In addition,our previous work found that the public awareness of MCI in Chuansha community was extremely low.In this study,cognitive assessment was conducted among 3,574 people aged ≥65in Chuansha Community,Shanghai,and the screened MCI patients were to know the prevalence of MCI,complete the baseline data examination.Aβ1-40,Aβ1-42,monocyte chemotactic protein-1(MCP-1),P-tau,APOE gene and diffusion tensor imaging(DTI),arterial spin labeling(ASL),diffusion weighted imaging(DWI)sequence examination were further completed to obtain fractional anisotropy(FA),cerebral blood flow(CBF)and apparent diffusion coefficient(ADC).All the participants experienced a two-year follow-up.Based on machine learning,a prediction model for MCI diagnosis and rapid deterioration of this community was established to analyze the characteristics of early warning for rapid progression of MCI.Chapter1 Cross-Sectional StudyPart 1 The Prevalence of MCI in the Elderly of Chuansha Community in ShanghaiObjective: To investigate the prevalence of MCI in the residents aged ≥65 in Chuansha Community of Shanghai.Methods: From June 20,2019 to December 30,2019,cognitive assessment and routine blood and urine biochemical tests were performed on the residents aged ≥65 in Chuansha Community,Shanghai.Results: A total of 3,574 patients aged ≥65 received Mini-cog Scale initial screening and Montreal Cognitive Assessment Scale(Mo CA).With various cognition-related diseases excluded,the results showed that 116 cases as the AD group,accounted for 5.7%;that 245 cases as the MCI group,12.2%;and that 1,651 cases as the normal group,82.1%.Sixteen hematologic indexes,including highdensity lipoprotein and red blood cell count et al,were significantly different among the three groups(P<0.05).Multiple regression analysis showed that years of education were negatively correlated with the degree of cognitive impairment,while age was positively(P<0.05).Conclusion: The prevalence of MCI in Chuansha Community of Shanghai was 12.2%.Insufficient education and advanced age were independent risk factors for cognitive impairment.Part 2 The Analysis of Serum Aβ,P-tau,MCP-1and APOE Genotype Correlated with AD and MCIObjective: To investigate the serum levels of Aβ1-42,Aβ1-40,Aβ1-42/Aβ1-40,P-tau181,MCP-1 and APOE genotype correlated with AD and MCI.Methods: Among 3574 screening population,a total of 297 individuals aged ≥65 were enrolled,including 68 patients in the AD group,105 in the MCI group and 124 in the normal group.The serum levels of Aβ1-40,Aβ1-42,P-tau181 and MCP-1 were determined by enzyme-linked immunosorbent assay(ELISA),APOE genotype was detected by Sanger sequencing.Results: The APOE4 genotype and serum Aβ1-42 level in AD group were significantly higher than those in MCI group,and those in MCI group were significantly higher than those in normal group(P<0.001);The levels of Aβ1-40,Aβ1-42/Aβ1-40,P-tau181 and MCP-1 were not significantly different among the three groups(P>0.05).Regression analysis suggested that APOE4 genotype and elevated serum Aβ1-42 level were independent risk factors for AD(P<0.05),while Aβ1-40,Aβ1-42/Aβ1-40,P-tau181 and MCP-1 had no correlation with AD(P>0.05);APOE4 genotype was an independent risk factor for MCI(P<0.001),while other hematological indicators were not correlated with MCI(P>0.05).Serum Aβ1-42 levels were negatively correlated with cognitive domain scores of visuospatial executive ability(P<0.05).Conclusions: The elevated serum Aβ1-42 level was an independent risk factor for AD,and was significantly associated with impaired visual spatial executive function,while APOE4 genotype was a common risk gene for AD and MCI.Part 3 The Analysis of Multimodal MRI Features correlated with MCI Based on Brain Image SegmentationObjective: To explore the multimodal MRI features of MCI individuals aged ≥65 in Chuansha Community of Shanghai.Methods: Among 3574 screening population,105 MCI patients(MCI group)and 124 normal subjects(NC group)underwent multimodal cranial MRI sequences of DTI,ASL,and DWI in 2019.The brain image was segmented into 116 regions by brain image segmentation technology,and the ADC value,FA value and CBF value of each brain region were extracted by deep learning algorithm.The imaging characteristics of the two groups were analyzed and compared.Results: Comparison of multi-modal MRI features between MCI group and normal group showed that: Among 116 brain regions,CBF values in 22 brain regions(Left superior frontal gyrus-dorsolateral,Right superior frontal gyrus-dorsolateral,Right inferior frontal gyrus,opercular part,Right rolandic operculum,Left olfactory cortex,Right olfactory cortex,Left insula,Right insula,Left anterior cingulate and paracingulate gyri,Right anterior cingulate and paracingulate gyri,Left median cingulate and paracingulate gyri,Right median cingulate and paracingulate gyri,Left parahippocampal gyrus,Left angular gyrus,Left precuneus,Left thalamus,Right thalamus,Left heschl gyrus,Right heschl gyrus,Left cerebelum＿3,Left cerebelum＿8,Vermis＿1＿2)were significantly different between the two groups,and the MCI group was significantly lower than the normal group(P<0.05);FA values in 15 brain regions(Left superior frontal gyrus-dorsolateral,Right superior frontal gyrus-dorsolateral,Left middle frontal gyrus,Right middle frontal gyrus,Left middle frontal gyrus-orbital part,Right middle frontal gyrus-orbital part,Left superior parietal gyrus,Left inferior temporal gyrus,Left cerebelum＿4＿5,Left cerebelum＿8,Left cerebelum＿9,Right cerebelum＿9,Left cerebelum＿10,Vermis＿8,Vermis＿10)were significantly different between the two groups,and the MCI group was significantly lower than the normal group(P<0.05).There was no significant difference in ADC values between the two groups(P>0.05).Conclusion: The decrease of CBF in 22 brain regions and FA in 15 brain regions may be a warning feature of MCI.Chapter2 Cohort StudyMultimodal MRI Features of Community Deteriorative MCI Based on Brain Image SegmentationObjective: To explore the multimodal MRI features of deteriorative MCI individuals aged ≥65 in Chuansha Community of Shanghai.Methods: Among 3574 screening population,a number of 105 MCI-afflicted patients underwent multimodal cranial MRI sequences of DTI,ASL,and DWI in 2019 and 2021,respectively.The brain image was segmented into 116 regions using brain image segmentation technology,and the ADC value,FA value and CBF value of each brain region were extracted by deep learning algorithm.The cognitive function of all MCI patients was followed up for 2 years,and they were divided into deteriorative MCI group and non-deteriorative MCI group according to the evaluation results.The imaging characteristics of the two groups and each group in two years were analyzed respectively.Results: Among the 105 MCI patients,41 deteriorated(deterioration group,DG)and 64 did not deteriorated(non-deterioration group,NDG)within 2 years.(1)As to CBF value,we observed the characteristics in 116 brain regions as follows:In 2019,the CBF value of DG was significantly higher than that of NDG(P<0.001),there was no significant difference between the two groups in 2021(P>0.05),the CBF values of each group in 2021 were significantly lower than those in 2019(P<0.001).(2)With FA values analyzed,we found the characteristics in nine brain regions(Middle frontal gyrus-orbital part,Left middle occipital gyrus,Right middle occipital gyrus,Right Middle temporal gyrus,Left superior temporal gyrus,Right superior temporal gyrus,Left cerebelum＿7b,Right cerebelum＿7b,Left cerebelum crura)as follows: In both 2019 and 2021,the FA value was significantly lower in DG than in the NDG(P<0.001).In the DG,FA value was significantly lower in 2021 than in 2019(P<0.001),while in the NDG,no significant difference was found in FA value between 2019 and 2021(P>0.05).(3)In terms of ADC value,we found the characteristics in Middle frontal gyrus orbital part region as follows: In both 2019 and 2021,the ADC value was significantly higher in the DG than in the NDG(P<0.001).In the DG,the ADC value was significantly higher in 2021 than in 2019(P<0.001).Conclusion: The early transient elevation of CBF in 116 brain regions,the decrease of FA in 9 brain regions and the increase of ADC in 1 brain region could be the warning characteristics of the deterioration of MCI.Chapter3 Application of Machine LearningPart 1 Machine-based Learning Shifting to Community Diagnostic Model of MCIObjective: To establish a diagnostic model of MCI by analyzing the features of individuals aged ≥65 in Chaunsha community of Shanghai based on the method of machine learning random forest method.Methods: Among 3574 screened subjects,229 case(105 cases in MCI group and 124 cases in normal group)completed gender,age and years of education(hereinafter referred to as "demographic characteristics"),and tested serum Aβ1-40,Aβ1-42,P-tau181 and MCP-1 levels,APOE gene(hereinafter referred to as " hematological indicators")and examined head MRI and extracted data(the images were divided into 116 brain regions by brain image segmentation technology,and ADC,FA and CBF values of each brain region were extracted by deep learning algorithm),and 357 features were collected.Based on Python platform Anaconda,the random forest algorithm was used to analyze all features,and 229 patients were randomly divided into training set(70%)and test set(30%)to establish MCI diagnostic model.Results:(1)Model 1 was based on 357 global features,the data of which showed that the accuracy of training set was 100%,the accuracy of test set was 100%,the sensitivity was 100%,the specificity was 100%,AUC = 1,the first fifteen features were FA value of left cerebelum＿4＿5,FA value of left Cerebelum＿8,FA value of left olfactory cortex,FA value of vermis＿8,Aβ1-42,ADC value of left cerebelum＿8,FA value of left cerebelum＿9,FA value of right inferior temporal gyrus,P-tau181,ADC value of right posterior cingulate gyrus,FA value of left calcarine fissure asurrounding cortex,FA value of vermis＿7,CBF value of left middle frontal gyrus,orbital part,CBF value of left inferior frontal gyrus,opercular part,CBF value of vermis＿3.(2)Model 2 was modeled again based on the first fifteen features of Model 1.The accuracy of training set was 100%,the accuracy of test set was 99%,the sensitivity was 97%,and the specificity was 100%,with AUC=1.(3)Model 3 was modeled again based on the first ten features of Model 1.The accuracy of training set was 100%,the accuracy of test set was 97%,the sensitivity was 95%,the specificity was 100%,and AUC was 0.97.(4)Model 4 was based on multi-modal MRI image features,the accuracy of training set was 100%,the accuracy of test set was 100%,the sensitivity was 100%,and the specificity was 100%.AUC=1,the first ten features were FA value of left cerebellum 4＿5,FA value of left cerebellum＿8,FA value of vermis＿8,FA value of left cerebellum＿9,FA value of left olfactory cortex,ADC value of left cerebellum＿8,CBF value of left cerebellum＿3,ADC value of the right posterior cingulate gyrus,FA value of the left calcarine fissure asurrounding cortex and CBF value of left cerebellum＿8.(5)Model 5 was modeled based on the first ten features of Model 4,with training set was 100%,the accuracy of test set was 100%,the sensitivity was 100%,and the specificity was 100%,with AUC=1.(6)The models based on demographic characteristics and hematological indicators respectively have low sensitivity and specificity.Conclusion: Model 5 based on ten features of multimodal MRI can be used for community MCI diagnosis with high sensitivity and specificity.Part 2 Machine-based Learning Shifting to Community Prediction Model of Deteriorative MCIObjective: To establish a prediction model of deteriorative MCI by analyzing the features of individuals aged ≥65 in Chaunsha community of Shanghai based on the method of machine learning random forest method.Methods: A total of 105 MCI individuals aged ≥65 were followed up,with a collection of 357 features,which were derived from the demographic characteristics,hematological indicators(serum Aβ1-40,Aβ1-42,P-tau181 and MCP-1 levels,APOE gene),and multimodal brain MRI imaging indicators of 116 brain regions(ADC,FA and CBF values).Cognitive function was followed up for 2 years.Based on Python platform Anaconda,105 patients were randomly divided into training set(70%)and test set(30%)by analyzing all features through random forest algorithm,and a prediction model of deteriorative MCI was established.Results:(1)model 1 was established based on demographic characteristics,hematological indicators and multi-modal MRI image features,the accuracy of training set was 100%,the accuracy of test set was 64%,the sensitivity was 50%,the specificity was 67%,and the AUC = 0.72.The first five features were APOE4 gene,FA value of left fusiform gyrus,FA value of left inferior temporal gyrus,FA value of left parahippocampal gyrus and ADC value of right talar gyrus.(2)Model 2 was modeled again based on the first five features of Model 1 with training set 100%,the accuracy of test set 85%,the sensitivity 91%,the specificity 80%,and the AUC was 0.96.(3)Model 3 was based on the first four features of Model 1 with 100% accuracy of training set,97% accuracy of test set,100% sensitivity,95% specificity,and AUC=0.99.(4)Model 4 was modeled again based on the first three characteristics of Model 1 with training set 100%,the accuracy of test set 94%,the sensitivity 92%,the specificity 94%,and the AUC was 0.96.(5)Based on hematological characteristics,model 5 showed 100% accuracy of training set,91% accuracy of test set,100% sensitivity,88% specificity,AUC=0.97.(6)The other models were respectivly based on demographic characteristics,imaging characteristics FA,CBF and ADC values have low sensitivity and specificity.Conclusion: Model 3 can be used to predict the deterioration of MCI in the community with high sensitivity and specificity,with four important predictive characteristics(APOE4 genotype and FA values of three brain regions).