Enhancing Alzheimer's Disease Classification with Transfer Learning: Finetuning a Pre-trained Algorithm.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neurology

Background:

• Alzheimer's disease (AD) poses a significant public health challenge due to increasing longevity.
• Accurate classification of AD stages is crucial but difficult due to intra-stage variability and manual classification errors.
• Deep learning models, such as ResNet50V2, show promise for improving image classification tasks.

Purpose of the Study:

• To leverage the ResNet50V2 model for accurate Alzheimer's disease classification using MRI scans.
• To investigate the impact of input layer sizes and fine-tuning strategies on model performance for AD staging.
• To develop a precise and reliable method for classifying Alzheimer's disease stages.

Main Methods:

• A dataset of 6400 verified MRI images was utilized.
• Transfer learning and fine-tuning of the ResNet50V2 model were employed for multi-class AD classification.
• The study explored various input layer sizes and optimal layer unfreezing strategies, incorporating custom layers and dynamic learning rate reduction.

Main Results:

• Model performance was evaluated using accuracy, AUC, precision, recall, F1-score, and ROC curves.
• Confusion matrices visualized model behavior, aiding in understanding classification performance.
• The developed model achieved a high accuracy of 96.18% in discriminating between AD stages.

Conclusions:

• The deep learning approach, specifically using a fine-tuned ResNet50V2 model, significantly enhances accuracy and reliability in Alzheimer's disease classification.
• This method holds promise for real-world clinical applications in diagnosing and staging Alzheimer's disease.
• The study highlights the potential of advanced AI techniques in addressing complex medical diagnostic challenges.