Early detection of Alzheimer's disease via multimodal MRI and machine learning

This study aimed to identify effective biomarkers associated with early-stage Alzheimer's disease (AD) by integrating multimodal neuroimaging features with machine learning (ML), addressing clinical challenges posed by global population aging.

Multimodal neuroimaging-including resting-state functional MRI (rs-fMRI), structural MRI (sMRI), and diffusion tensor imaging (DTI)-was combined with ML techniques. A total of 234 subjects [cognitively normal (CN), mild cognitive impairment (MCI), and AD] were selected from the AD Neuroimaging Initiative (ADNI) database. Brain functional, structural, and microstructural features were extracted, and nine ML models, including support vector machine (SVM), random forest (RF), and Naive Bayes, were trained and evaluated across three binary classification tasks: AD-CN, MCI-CN, and AD-MCI.

The SVM model achieved the highest performance for AD-CN (AUC = 0.901) and MCI-CN (AUC = 0.839), while RF performed best for AD-MCI classification (AUC = 0.809). Functional analyses revealed significant abnormalities in key regions, including the anterior cingulate cortex, hippocampus, and middle frontal gyrus in AD patients. Structural analyses confirmed that hippocampal subfield atrophy was strongly associated with cognitive decline. Diffusion metrics, particularly the DTI-ALPS index, reflected microstructural white matter damage effectively.

Integrating multimodal neuroimaging with ML enhances diagnostic accuracy for AD and MCI and identifies potential neuroimaging biomarkers, providing objective evidence to support early clinical intervention.