A comparative study of early stage Alzheimer's disease classification using various transfer learning CNN frameworks

Area of Science:

• Artificial Intelligence
• Medical Imaging Analysis
• Neuroscience

Background:

• Machine learning and deep learning offer improved predictive performance and computational efficiency.
• Transfer learning within Convolutional Neural Networks (CNNs) is effective for diagnosing Alzheimer's disease stages.

Purpose of the Study:

• To enhance existing CNN architectures (Xception, InceptionResNetV2, DenseNet201, InceptionV3, ResNet50, MobileNetV2) for Alzheimer's disease diagnosis.
• To improve model effectiveness and precision by incorporating batch normalization, dropout, and dense layers.

Main Methods:

• Utilized transfer learning with base CNN architectures.
• Extended models with batch normalization, dropout, and dense layers.
• Evaluated models on Kaggle MRI Alzheimer's dataset (5120 training, 1280 testing images) using precision, recall, F1-score, accuracy, and ROC-AUC metrics.

Main Results:

• The enhanced Xception model achieved 99% accuracy and a 0.135 loss score without K-fold cross-validation.
• With five K-fold cross-validation, accuracy increased to 99.68% and loss decreased to 0.120.
• ROC-AUC evaluation further supported the model's diagnostic capabilities.

Conclusions:

• The enhanced Xception model demonstrates high accuracy and efficiency for Alzheimer's disease detection and classification.
• The study validates the effectiveness of transfer learning and model augmentation for medical image analysis.
• The proposed approach offers a promising tool for rapid and accurate Alzheimer's disease diagnosis.