Deep learning based pipelines for Alzheimer's disease diagnosis: A comparative study and a novel deep-ensemble method

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Medical Diagnostics

Background:

• Alzheimer's disease is a neurodegenerative condition with no cure, impacting cognitive functions.
• Deep learning methods are increasingly used in neuroimaging for Alzheimer's diagnosis due to their ability to process raw data objectively.
• Training reliable deep learning models is challenging due to variations in brain image types.

Purpose of the Study:

• To develop and compare deep learning models for computer-aided diagnosis (CAD) of Alzheimer's disease.
• To present a fully automated deep-ensemble approach for classifying dementia levels from brain images.
• To evaluate the robustness of the deep learning strategy for detecting Alzheimer's disease and different dementia stages.

Main Methods:

• A deep-ensemble approach was developed and tested on heterogeneous MRI and fMRI datasets.
• The strategy involved comprehensive analysis for both binary (Alzheimer's vs. non-Alzheimer's) and multiclass (dementia levels) classification.
• Deep learning architectures were compared to identify the most suitable for the task.

Main Results:

• The proposed deep-ensemble approach achieved high accuracy, exceeding state-of-the-art results.
• Achieved 98.51% accuracy in binary classification and 98.67% in multiclass classification, averaged across four datasets.
• Demonstrated robustness and reliability through cross-dataset experiments.

Conclusions:

• The deep-ensemble approach offers a robust and reliable CAD system for Alzheimer's disease detection and dementia-level classification.
• The strategy can be extended to other imaging techniques beyond MRI and fMRI.
• This approach has considerable potential to benefit patient management in clinical settings.