| PART Ⅰ.The Spatial Distribution of Brain Function Activity Changes in Chronic Kidney Disease Patients:A Resting-State Functional Magnetic Resonance Imaging StudyObjective:To explore the spatial distribution of brain function activity changes in chronic kidney disease(CKD)patients based on whole brain voxel-level analysis.Materials and methods:This study retrospectively recruited 621 CKD patients and 625 age,sex and education level matched healthy controls.All subjects completed neuropsychological assessments and brain MRI scan.According to inclusion and exclusion criteria,380 CKD patients and 404 healthy controls were enrolled in this study.CKD patients were further divided into end stage renal disease(ESRD)group and non-ESRD group using 1:1 propensity score matching(PSM)according to the kidney function of each subject.Amplitude of low-frequency fluctuation(ALFF),fractional amplitude of low-frequency fluctuation(fALFF),regional homogeneity(ReHo),binarized degree centrality(BDC)and weighted degree centrality(WDC)were calculated based on resting-state functional magnetic resonance imaging(rs-fMRI).These whole brain voxel-level brain functional activity indicators were then compared between each group.Further,independent component analysis(ICA)was applied to determine the spatial distribution of each sub-network,and the number and proportion of voxels based on aberrant brain regions were calculated to summarize the general rule of spatial distribution characteristics of brain functional activity changes in CKD patients.Results:The neuropsychological assessments showed that the score of number connecting test(NCT),line tracing test(LTT)and serial dotting test(SDT)were significantly higher in the CKD group,while the socre of digit symbol test(DST)was significantly lower in the CKD group(all p<0.05).In the subgroup analysis,the score of NCT was significantly higher in the ESRD group,while the socre of DST was significantly lower in the ESRD group(all p<0.05).Whole brain voxel-level analysis showed that CKD patients showed a wide range of aberrant functional activity(mostly reduced functional activity),which mainly located in the frontal and parietal lobes.A few brain regions showed increased functional activity,mostly located in the temporal and occipital lobes.Similar findings were found in sub-group analysis.The voxel number and voxel proportion analysis showed that the abnormal functional activity mainly located in the default mode network(DMN),frontal parietal network(FPN)and sensorimotor network(SMN),and the DMN was the most prominent network.Conclusion:Based on large sample neuroimaging data,this study systematically explored the spatial distribution patterns of brain functional activity changes in CKD patients,and confirmed that the DMN,FPN and SMN are the most prominent impaired regions in CKD patients,especially the DMN.PART Ⅱ.Cognitive Impairment Related Brain Network Changes in Chronic Kidney Disease Patients:A Graph Theory-based Magnetic Resonance Imaging StudyObjective:To explore the cognitive impairment related brain network changes in chronic kidney disease(CKD)patients using graph theory-based analysis,and the relationship between brain network topological properties and cognitive function.Materials and methods:This study retrospectively recruited 621 CKD patients and 625 age,sex and education level matched healthy controls.All subjects completed neuropsychological assessments and brain MRI scan.According to inclusion and exclusion criteria,380 CKD patients and 404 healthy controls were enrolled in this study.CKD patients were further divided into end stage renal disease(ESRD)group and non-ESRD group using 1:1 propensity score matching(PSM)according to the kidney function of each subject.Graph theory was used to analyze network topological properties of DMN,FPN and SMN.These network indicators(global properties,nodal properties and functional connectivity)were then compared between each group.Correlation analysis was also applied between these network properties and neuropsychological test scores as well as laboratory test results.Results:The graph theory analysis showed that the CKD patients showed obviously functional network impairment in both three sub-network compared with control group.In the global properties,CKD patients showed decreased global efficiency,local efficiency and clustering coefficient and increased shortest path.In the nodal properties,CKD patients showed decreased nodal degree,nodal clustering coefficient,nodal global efficiency and nodal local efficiency and increased nodal betweeness and nodal shortest path.In the functional connectivity,CKD patients showed extensive functional connectivity impairment(mostly decreased and partly compensated increased).Similar findings were found in sub-group analysis,while the alteration in FPN is the least obvious.The correlation analysis showed that brain network topological properties were significantly correlated with neuropsychological test scores as well as levels of nephrotoxic substances.Conclusion:Based on the advanced graph theory analysis algorithm,this study found that CKD patients showed abnormal brain network topological properties of default mode network,frontal parietal network and sensorimotor network,and these changes correlated with the cognitive performance and clinical indicators of patients.These results revealed the neural mechanism of cognitive impairment in CKD in depth.PART Ⅲ.Predicting Cognitive Impairment in Chronic Kidney Disease Patients Using Structural and Functional Brain Network:An Application Study of Machine LearningObjective:To investigate whether a machine learning model based on structural and functional brain network could predict cognitive impairment in chronic kidney disease(CKD)patients.Materials and methods:This study retrospectively recruited 380 CKD patients in Jinling hospital and 54 CKD patients in another hospital.These patients were further divided into cognitive impairment(CI)group and non-CI group based on diagnostic criteria.All patients underwent brain MRI scan,neuropsychological test and laboratory exam.A deep learning model(attentionMLP)was trained and applied for the discrimination of CI based on multimodality network topological properties.Finally,a logistic regression model was built combining MRI features and clinical features.The area under curve(AUC),sensitivity and specificity were calculated to evaluate the model performance.Delong test was used to examine the difference of AUCs between models.The integrated discrimination improvement(IDI)and net reclassification index(NRI)between models were calculated.Results:AttentionMLP model performed well in both internal test set and external test set(AUC=0.744 and 0.763,respectively).The AUC of the attentionMLP model was significantly higher than that of traditional machine learning model(logistic regression:p=0.004;support vector machine:p=0.005;decision tree:p=0.002;XGBoost:p=0.005;respectively,Delong test)in internal test set.Model performance was further improved in both the internal(AUC:0.748)and external test sets(AUC:0.774)after combining clinical features,and both IDI and NRI results showed that the combination of clinical features significantly improved the classification performance of the model in the external test set(all p<0.05).For the model using different modality data,the IDI results showed that compared to using single modality data,combining structural and functional modality data significantly improved the classification performance of the models in the internal test set(all p<0.05).For the model using different forms of data,the IDI results showed that using the brain network topological property data significantly improved the classification performance of the model,both in the internal test set and the external test set,compared to using the raw connection matrix data(all p<0.05).Conclusion:The result indicated that the deep-learning model could help predict the CI of CKD patients effectively based on structural and functional network topological properties. |
