A systematic analysis of diagnostic performance for Alzheimer's disease using structural MRI

Area of Science:

• Neuroimaging
• Medical Informatics
• Geriatric Medicine

Background:

• Alzheimer's disease (AD) is a prevalent neurodegenerative disorder in the elderly population.
• Structural magnetic resonance imaging (sMRI) studies have achieved high accuracy (80-95%) in distinguishing AD from normal controls (NCs).
• Clinical translation of these diagnostic models into practical computer-aided diagnosis (CAD) systems for AD remains limited.

Purpose of the Study:

• To identify factors impeding the clinical adoption of sMRI-based diagnostic models for Alzheimer's disease.
• To systematically review and analyze existing diagnostic models for AD using sMRI.

Main Methods:

• A systematic literature review of sMRI-based AD diagnostic models published in the last 15 years.
• Inclusion of 101 studies based on predefined screening criteria.
• Evaluation of heterogeneity and publication bias using subgroup analysis, meta-regression, and statistical tests (Begg's, Egger's).

Main Results:

• Recently published studies show high diagnostic accuracy for AD detection using sMRI.
• Significant heterogeneity was observed across studies, attributed to diverse datasets, machine learning models (traditional vs. deep learning), cross-validation techniques, sample sizes, and publication dates.
• Cross-validation methods, such as k-fold, can overestimate accuracy compared to independent sample validation.

Conclusions:

• While sMRI-based models demonstrate potential for classifying AD from NC, considerable heterogeneity exists.
• Factors such as varied methodologies and data sources present challenges for developing clinically applicable early AD diagnostic systems.
• Further research and standardization are necessary to bridge the gap between research findings and clinical practice in AD diagnosis.