Cogninet: an explainable deep learning model for multi-class MRI-based Alzheimer's disease staging.

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Medical Diagnostics

Background:

• Alzheimer's Disease (AD) diagnosis faces challenges in early detection and clinical decision-making.
• Current research often uses binary classification and lacks model interpretability and clinical usability.

Purpose of the Study:

• To introduce CogniNet, a novel deep learning model for Alzheimer's progression classification.
• To address limitations in existing AI models for AD diagnosis, focusing on interpretability and clinical utility.

Main Methods:

• Developed CogniNet, a convolutional neural network (CNN) combining VGGNet19 and DenseNet201 architectures.
• Utilized T1-weighted MRI scans for four-way classification of Alzheimer's progression.
• Employed Grad-CAM for generating attention maps to enhance model interpretability.

Main Results:

• CogniNet achieved 98% accuracy and 98% sensitivity on 3,200 unseen MRI slices.
• Demonstrated superior performance compared to established CNN architectures and prior research.
• Grad-CAM provided visual explanations for model predictions.

Conclusions:

• CogniNet is a high-performing, explainable deep learning model for AI-assisted neuroimaging diagnostics.
• Interpretable outputs from Grad-CAM foster clinical trust and adoption.
• The model shows promise for improving Alzheimer's Disease diagnosis and management.