Pre-training and ensembling based Alzheimer's disease detection.

Area of Science:

• Artificial Intelligence in Medicine
• Neuroimaging and Signal Processing
• Geriatric Health and Disease Detection

Background:

• Alzheimer's disease (AD) poses a significant health challenge for the elderly.
• Early diagnosis, particularly during the Mild Cognitive Impairment (MCI) phase, is critical due to limited effective treatments.
• Objective diagnostic tools are needed to support clinical decision-making.

Purpose of the Study:

• To develop an automated classification technology for Alzheimer's disease detection.
• To enhance Alzheimer's disease detection using artificial intelligence, aiming to reduce diagnostic costs.
• To evaluate a novel pre-trained ensemble-based AD detection (PEADD) framework.

Main Methods:

• Proposed a novel pre-trained ensemble-based AD detection (PEADD) framework utilizing ResNet, VGG, and EfficientNet base learners.
• Investigated context-enriched image modalities for audio-based AD detection and employed image denoising strategies.
• Implemented PET (Positron Emission Tomography)-based AD detection using denoised PET images and explored hard and soft voting ensemble methods.

Main Results:

• Achieved classification accuracies of 92% for audio-based and 99% for PET-based AD detection.
• The PEADD framework demonstrated superior performance compared to state-of-the-art methods on both datasets.
• Validated the effectiveness of context-enriched audio analysis and denoised PET imaging for AD detection.

Conclusions:

• The developed network model offers an objective basis for healthcare professionals in detecting Alzheimer's disease.
• The PEADD framework shows significant potential for improving early and accurate AD diagnosis.