A Support Vector Machine Method With Correlation-Based Recursive Feature Elimination for the Early Diagnosis of Alzheimer's Disease

Alzheimer's disease (AD) is a progressive neurodegenerative disorder where early diagnosis is critical. While magnetic resonance imaging (MRI) provides valuable structural information, most methods rely solely on imaging features, limiting clinical interpretability.

A novel feature selection method using support vector machine with correlation factor-based recursive feature elimination and cross-validation (SVM_CF_RFECV) integrated correlation factors from mini-mental state examination (MMSE) scores into the feature weighting process is proposed for the early identification of Alzheimer's disease. This approach balances statistical discriminative power with clinical relevance and mitigates the local-optimum issue of traditional recursive feature elimination (RFE).

Using the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset, the proposed method achieved a peak accuracy of 87.5% in a single-split experiment and an average accuracy of 0.8489 ± 0.0330 across multiple random partitions, outperforming conventional SVM_RFECV. The selected 74 discriminative brain regions, including the hippocampus and cingulate cortex, align with established AD pathology.

These findings demonstrate that SVM_CF_RFECV provides both accuracy and interpretability, supporting its potential as a clinically meaningful tool for early AD diagnosis.