Diagnosing Alzheimer's Disease Based on Multiclass MRI Scans using Transfer Learning Techniques.

Area of Science:

• Medical image analysis
• Artificial intelligence in healthcare
• Neurological disorder diagnostics

Background:

• Early prediction and classification of Alzheimer's disease (AD) are critical for preventing progression to dementia.
• Medical image analysis plays a crucial role in the early detection of AD.
• Deep learning (DL) models offer advanced capabilities for computer-assisted diagnosis.

Purpose of the Study:

• To employ a deep learning (DL) model for the early detection and classification of Alzheimer's disease (AD).
• To utilize a transfer learning technique for classifying magnetic resonance (MR) images.
• To compare the effectiveness of different convolutional neural network (CNN) base models in identifying AD.

Main Methods:

• Classified Alzheimer's disease (AD), mild cognitive impairment (MCI), and normal control (NC) subjects using 2D MR images.
• Utilized state-of-the-art CNN base models (ResNet-101, ResNet-50, ResNet-18) for image classification.
• Evaluated models on the AD Neuroimaging Initiative (ADNI) dataset, focusing on hippocampus regions.

Main Results:

• ResNet-101 achieved 98.37% accuracy, outperforming ResNet-50 and ResNet-18 in multiclass classification.
• The study demonstrated the effectiveness of transfer learning, particularly with limited datasets.
• The models showed strong performance in distinguishing between AD, MCI, and NC subjects.

Conclusions:

• Transfer learning effectively addresses deep learning challenges associated with insufficient data for training models from scratch.
• This approach is highly advantageous for medical image analysis in diagnosing diseases like AD.
• The findings highlight the potential of CNN-based transfer learning for accurate and efficient AD diagnosis.