Structural Neuroimaging Markers of Dementia: Insights from ROC Curve Analysis

Background Alzheimer’s disease and related dementias (ADRD) are growing public health concerns. Early and accurate differentiation between cognitively normal (NL), mild cognitive impairment (MCI), and dementia is essential for timely intervention. Structural MRI biomarkers—such as grey matter volume (GMV), white matter volume (WMV), white matter hyperintensity volume (WMH), AD signature (a meta-region of interest), and hippocampal occupancy score (HOC)—offer promise for objectively staging cognitive decline. Objective This study aimed to evaluate the diagnostic utility of MRI-derived neuroimaging measures using receiver operating characteristic (ROC) curve analysis to identify optimal thresholds for distinguishing between NL, MCI, and dementia. Methods Data were collected from 466 participants at the KU Alzheimer’s Disease Research Center between 2010 and 2025. T1-weighted MRI scans were preprocessed using SPM12 and VBM12. Statistical analyses included ANOVA, Tukey’s HSD, Welch’s t-tests, and ROC analyses. Youden’s J statistic was used to determine optimal cut-points, and multivariate models assessed combined biomarker performance. Results Significant differences were found across diagnostic groups for GMV, WMH, AD signature, and HOC. ROC analysis for Demented vs the control group (NL) showed GMV had an AUC of 0.674 [0.6164, 0.7325], WMH an AUC of 0.700 [0.6454, 0.7539], AD signature an AUC of 0.829 [0.780, 0.8761], and HOC the highest individual performance with an AUC of 0.878 [0.8426, 0.9144]. Further ROC analysis confirmed AD signature and HOC as the most effective biomarkers (AUCs = 0.78 and 0.81) for distinguishing MCI/demented from NL. In MCI vs. demented comparisons, AD signature and HOC biomarkers had AUCs of 0.70 and 0.73, respectively. Conclusions MRI-derived biomarkers, particularly AD signature and HOC, show strong potential for differentiating cognitive impairment stages. ROC-based thresholds offer clinically actionable metrics that can improve diagnostic precision.