Brain connectome gradient dysfunction in patients with end-stage renal disease and its association with clinical phenotype and cognitive deficits

Figures

• 

  Fig. 1. Flowchart demonstrating the phases in patients with end-stage renal disease from screening to inclusion and completion of the study.

• 

  Fig. 2. Connectome gradient mapping in ESRD patients and HC. a Gradient 1 was organized along a gradual axis from VIS/SMN to DMN. b, c Gradient 2 and gradient 3 extended between the SMN and the VIS. a–c Regions with similar connectivity patterns demonstrate similar coloring. The gradients were overall similar between-group, while lower/higher values in ESRD patients are presented as blue/red. The statistical significance level was set as P values of < 0.05 after FDR correction. d–f Global histograms of the gradients 1–3 showing contracted extremes in ESRD patients relative to the HC range, whereas those of gradient 1 and gradient 3 in the middle axis increased and gradient 2 in the midrange decreased. g–l. System-based analysis of the gradient score between-group. VIS visual network, SMN sensorimotor network, DAN dorsal attention network, VAN ventral attention network, LIM limbic network, FPCN frontoparietal network, DMN default mode network.

• 

  Fig. 3. Statistical comparison of gradient metrics. a Between-group differences in global metrics of the first three gradients. b Between-group differences in gradient scores and correlations with clinical and cognitive variables across all patients. c Word clouds of cognitive functions associated with brain regions showing lower (blue) or higher (red) between-group gradient scores. The resulting font size of the cognitive term corresponds to the correlation coefficient (r) of the between-group Z-map of the first three gradients and the meta-analytic map of that word generated by Neurosynth. All correlations were significant (P < 0.05, FDR corrected).

• 

  Fig. 4. Performance of machine learning algorithms for classification. a ROC curve of the classifiers. The accuracy and AUC of logistic regression were 84.8% and 0.901, respectively, which was better than those of the other classifier algorithms. b Features correlation heatmap. ROC receiver operating characteristic, AUC area under the curve, KNN k-nearest neighbor, SVM support vector machine, DT decision tree, RF random forest, GBDT gradient boosting decision tree.