Early Diagnosis of Alzheimer's: Machine Learning Analysis Leveraging Structural MRI

Alzheimer's disease (AD) is characterized by significant brain atrophy, detectable via structural MRI. There has been less focus on cortical degeneration in subcortical regional deterioration-particularly during the transition from Mild Cognitive Impairment (MCI) to AD-which remains underexplored. This study aims to identify subcortical regions with progressive atrophy using surface-based morphometry (SBM) and evaluate their potential for early AD diagnosis.

This longitudinal study collected data from MCI patients (6 months) who later progressed to Alzheimer's within 3 years, alongside healthy controls followed across four time points (6 months-3 years). Reported in line with STARD guidelines, the study aimed to evaluate model performance in distinguishing progressive MCI-to-AD from healthy controls to advance early Alzheimer's diagnosis. The study leveraged the ADNI dataset to analyse 68 subcortical regions in MCI-to-AD converters (MCI-AD) and Healthy Controls (HC) over 6 months to 3 years (i.e., at 6 months, 1 year, 2 years, and 3 years). The groups were classified beforehand using the Clinical Dementia Rating and the Mini-Mental Status Examination scores. Accordingly, three surfacebased morphometry (SBM) metrics-cortical thickness (CTh), gyrification index (GI), and sulcal depth (SD)-were evaluated in the progressive MCI group (individuals with MCI who later converted to Alzheimer's disease) as well as in healthy controls, to quantify morphological changes. Finally, the morphological data were utilized to train and test machine learning models for classification and prediction.

Cortical regions exhibiting significant atrophy were identified using paired-samples ttests between 6 months and 3 years. In parallel, machine learning (ML) models were trained and tested on the same dataset to differentiate progressive MCI-to-AD cases from healthy controls across multiple time points (6 months, 1 year, 2 years, and 3 years), and subsequently to predict the progression to Alzheimer's disease. Certain evaluation metrics were considered for the classifier performance, i.e., Accuracy, F1-score, and AUC-ROC.

Sub-cortical SBM metrics, particularly CTh, are sensitive to early AD-related atrophy, offering potential biomarkers for disease progression. ML models trained on these features enable accurate classification, with performance peaking near diagnosis-highlighting their utility in early intervention.

SBM-derived subcortical atrophy patterns aid early AD detection and, when combined with ML, offer a scalable predictive framework. Among metrics, CTh showed the greatest decline in MCI-AD, followed by SD and GI. Progressive deterioration was observed in specific subcortical regions, accelerating near diagnosis. Classifiers achieved higher accuracy in distinguishing MCI-AD from HC at later time points, highlighting stage-specific shifts in feature importance.