A bilateral filtering-based image enhancement for Alzheimer disease classification using CNN.

Area of Science:

• Medical Imaging Analysis
• Artificial Intelligence in Healthcare
• Neuroscience

Background:

• Alzheimer's disease (AD) diagnosis relies on accurate classification of neuroimaging data.
• Image noise in magnetic resonance imaging (MRI) datasets can degrade the performance of deep learning models for AD classification.
• Developing robust deep learning models requires addressing data quality challenges like image noise.

Purpose of the Study:

• To develop an optimally performing convolutional neural network (CNN) for classifying Alzheimer's disease (AD) into three categories: mild cognitive impairment (MCI), normal controls (NC), and AD.
• To mitigate the impact of image noise on deep learning model performance by implementing an image enhancement scheme.
• To evaluate the efficacy of the proposed CNN model in accurate AD classification using MRI data.

Main Methods:

• An image enhancement algorithm combining histogram equalization and bilateral filtering was applied to reduce noise and improve MRI dataset quality.
• A CNN model with four convolutional layers and two hidden layers was designed for AD classification.
• The CNN model was trained and evaluated using a 10-fold cross-validation approach with a learning rate of 0.001 and 200 epochs.

Main Results:

• The proposed CNN model achieved an accuracy of 93.45% and an Area Under the Curve (AUC) of 0.99 for the three-class classification (MCI, NC, AD).
• The model demonstrated superior performance in binary classification tasks compared to existing methods.
• Specific accuracies for binary classifications included 94.39% (AD vs. NC), 94.92% (AD vs. MCI), and 95.62% (MCI vs. NC).

Conclusions:

• The developed CNN model, utilizing enhanced MRI data, effectively classifies Alzheimer's disease, mild cognitive impairment, and normal controls.
• The image enhancement technique significantly improved the quality of MRI data, leading to enhanced classification performance.
• The study highlights the potential of deep learning with pre-processed imaging data for accurate and reliable Alzheimer's disease diagnosis.