Prediction of multi-stage Alzheimer’s disease progression based on surface-based morphometry and machine learning

Background Early identification of individuals at high risk of progression from normal cognition (NC) to mild cognitive impairment (MCI) and from MCI to Alzheimer’s disease (AD) is crucial for timely intervention. However, predictive models that simultaneously address multiple transition stages along the AD continuum remain limited. Objective To develop and validate machine learning models and nomograms based on surface-based morphometry (SBM) features, neuropsychological assessments, and plasma biomarkers for predicting disease progression in early mild cognitive impairment (EMCI), late mild cognitive impairment (LMCI), and NC populations. Methods This longitudinal study included 406 participants from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) database. SBM features, including sulcal depth, gyrification index, and fractal dimension, were extracted from 3D T1-weighted MRI. Multiple machine learning classifiers (support vector machine, linear discriminant analysis, LASSO, and Gaussian process) were employed to construct predictive models. An R score generated from the optimal machine learning model was combined with clinical predictors via logistic regression to build the nomogram. Model performance was assessed using receiver operating characteristic (ROC) curves. Results The study included 406 participants: 150 with EMCI, 67 with LMCI, and 189 with NC. The conversion rates to the next stage were 20.7% for EMCI-to-AD, 49.3% for LMCI-to-AD, and 26.5% for NC-to-MCI. The SBM-based machine learning models yielded AUCs of 0.769, 0.750, and 0.640 for predicting conversion from EMCI to AD, LMCI to AD, and NC to MCI, respectively, in the validation set. The integrated nomogram incorporating the R score and clinical predictors significantly improved predictive performance, achieving validation AUCs of 0.91 (EMCI to AD), 0.74 (LMCI to AD), and 0.73 (NC to MCI). Independent predictors for the different transition stages included the R score, tau protein (P-tau181), Aβ42/Aβ40 ratio, age, and Montreal Cognitive Assessment (MoCA) score, with variation across stages. Conclusion Integrating the SBM-derived R score with clinical and biomarker data through a nomogram provides a clinically interpretable tool for predicting multi-stage AD progression. This strategy may facilitate early risk stratification and the development of individualized intervention plans.