PCA-Enhanced Deep Features for Alzheimer's Disease Stage Classification with EFMM

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neuroscience

Background:

• Alzheimer's disease (AD) diagnosis requires accurate methods for timely intervention.
• Current deep learning models for AD classification can be computationally intensive.
• High-dimensional features in MRI data pose challenges for efficient classification.

Purpose of the Study:

• To develop a lightweight hybrid framework for MRI-based Alzheimer's disease stage classification.
• To improve the efficiency and accuracy of AD diagnosis using deep learning and dimensionality reduction.
• To evaluate the performance of various machine learning classifiers on reduced MRI feature sets.

Main Methods:

• Deep features were extracted from MRI images using a pre-trained SqueezeNet model.
• Principal Component Analysis (PCA) was applied for dimensionality reduction, retaining 100 components.
• Classifiers including Enhanced Fuzzy Min-Max Neural Network (EFMM) were tested using stratified 5-fold cross-validation.

Main Results:

• The SqueezeNet-PCA-EFMM framework demonstrated high classification accuracy (97.19% on test set).
• EFMM classifier achieved superior performance, with high AUC values for all AD stages.
• Stratified 5-fold cross-validation confirmed the framework's robustness, yielding a mean accuracy of 98.38%.

Conclusions:

• The proposed SqueezeNet-PCA-EFMM framework is effective and efficient for Alzheimer's disease stage classification from MRI.
• The hybrid approach combines deep learning, PCA, and EFMM for high performance and robustness.
• The EFMM's incremental learning capability makes this framework suitable for future clinical decision support systems.