Subjective cognitive decline (SCD) can precede mild cognitive impairment (MCI), but progression is variable. This study sought to determine which individuals with SCD are most likely to convert to MCI and over what time frame, and to quantify the incremental value of MRI measures of regional white matter hyperintensity (WMH) burden and gray matter (GM) atrophy beyond demographic and cognitive predictors.
Data were drawn from four longitudinal cohorts: the Alzheimer’s Disease Neuroimaging Initiative (ADNI), the National Alzheimer’s Coordinating Center (NACC), CIMA-Q, and PREVENT-AD. The pooled sample comprised 1,352 participants who were classified as having SCD at baseline. Image processing and WMH segmentation pipelines used in the study are open source, and the lobar atlas for regional WMH metrics is publicly available. The manuscript reports that all datasets were de-identified and that appropriate ethical approvals and participant consents were obtained.
Three supervised machine learning algorithms were applied: logistic regression, random forest, and XGBoost. Models were trained to predict conversion from SCD to MCI at four discrete time horizons: 2 years, 3 years, 4 years, and 5 years after baseline.
Four nested feature sets were evaluated to determine incremental predictive value:
Comparing these sets allowed assessment of how much MRI or cognitive information improved prediction beyond routine clinical variables.
Model performance was summarized using Area Under the Curve (AUC) values. Reported peak AUCs for the feature sets at each horizon were:
These results indicate that adding MRI-derived regional WMH and GM volumes improved discriminative accuracy, particularly at longer prediction intervals. At the 2-year horizon, combining cognitive and MRI features yielded the best result, whereas at 3–5 years MRI additions to the base model produced larger AUC gains than cognitive measures alone.
Among MRI predictors, regional measures were differentially informative. The analysis identified parietal WMH as the most frequently selected MRI predictor across models. Regional GM volumes were also incorporated and contributed to improved prediction, though the source emphasizes the prominence of regional WMH metrics in longer-term forecasting.
The study concludes that a targeted set of regional MRI markers—particularly WMH and regional gray matter volumes—added substantial predictive value over demographic and clinical features for forecasting SCD-to-MCI conversion up to five years in advance. Cognitive tests (MoCA, Trail Making Test B) improved near-term (2-year) prediction when combined with MRI, but MRI features provided greater incremental benefit at 3–5 year horizons. These findings support the potential role of regional MRI metrics in risk stratification for individuals with SCD and in selecting candidates for monitoring or preventive interventions.
The manuscript reports that data were obtained from ADNI, NACC, CIMA-Q, and PREVENT-AD. Image processing and WMH segmentation code and the lobar atlas used for regional WMH derivation are publicly available via the authors’ repositories and linked resources. The authors declare no competing interests and state that all required ethical approvals and participant consents were obtained. Specifics of IRB approvals, data access procedures for CIMA-Q membership, and relevant repository links are provided in the source manuscript.