Early detection of Alzheimer's disease using MRI and 2D machine learning systems

: enhancing clinical diagnostics

Abstract

The early detection of Alzheimer’s disease (AD) is critical for effective intervention and management. However, traditional classification systems—typically limited to cognitively normal (CN), mild cognitive impairment (MCI), and AD stages—often fail to capture the clinical heterogeneity and progressive nature of cognitive decline. This study introduces a comprehensive machine learning (ML) framework to enhance AD classification using structural Magnetic Resonance Imaging (MRI). We implement a refined nine-stage system that not only subdivides MCI into Early (EMCI), transitional (MCI), and Late (LMCI) stages but also incorporates beta-amyloid biomarker status (positive/negative) for further differentiation. Our research analyzed T1-weighted MRI scans from 714 subjects from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) database, categorized into nine groups: AD, CN, Subjective Memory Complaints (SMC), and amyloid-stratified versions of EMCI, MCI, and LMCI. A robust preprocessing pipeline was applied, followed by feature extraction using Histogram and Gray-Level Co-occurrence Matrix (GLCM) methods. Pairwise classification was performed using a suite of ML classifiers belonging to Scikit-learn library and feature selection techniques. The framework demonstrated high classification accuracy across numerous stage comparisons, achieving standout results such as 97.8% for AD vs. EMCI-Negative and 95.5% for AD vs. CN. Notably, the model also effectively distinguished between adjacent MCI stages, like MCI-Negative vs. LMCI-Negative, with 94.9% accuracy. Our findings indicate that beta-amyloid status enhances discriminability and that Histogram and GLCM features provide complementary diagnostic information, with the coronal plane frequently yielding the most informative results. This multi-stage classification approach shows significant promise for improving the precision of early and differential AD diagnosis.

Date of Award	28 Oct 2025
Original language	English
Awarding Institution	Universidade Católica Portuguesa
Supervisor	Pedro Miguel Rodrigues (Supervisor) & Maria Inês Barbosa (Co-Supervisor)