Lightweight Deep Learning Models with Explainable AI for Early Alzheimer's Detection from Standard MRI Scans

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Medical Diagnostics

Background:

• Dementia, including Alzheimer's Disease (AD), affects millions globally, with diagnosis challenging in resource-limited settings.
• Current diagnostic methods for AD often rely on costly neuroimaging and specialist expertise, hindering early detection.
• Early diagnosis of AD is critical for managing symptoms and slowing disease progression.

Purpose of the Study:

• To develop and evaluate computationally efficient deep learning models for early Alzheimer's Disease detection.
• To address the challenge of accessible and timely AD diagnosis in clinical practice.
• To enhance the interpretability of AI models in neuroimaging for clinical trust.

Main Methods:

• Utilized lightweight deep learning models, MobileNetV2 and EfficientNetV2B0.
• Trained models on 2D structural magnetic resonance imaging (MRI) slices for early AD detection.
• Applied explainability methods (Grad-CAM++, Guided Grad-CAM++) for model interpretability.

Main Results:

• The EfficientNetV2B0 model achieved 88.0% mean accuracy in distinguishing between Cognitively Normal (CN), Early Mild Cognitive Impairment (EMCI), and Late Mild Cognitive Impairment (LMCI).
• The model demonstrated strong performance in a multi-class classification task.
• Explainability methods successfully visualized the anatomical regions influencing model predictions.

Conclusions:

• Developed an accessible and interpretable neuroimaging tool for early AD diagnosis.
• The proposed deep learning models can extend expert-level diagnostic capabilities to routine clinical settings.
• This approach facilitates earlier intervention and management of Alzheimer's Disease.