Neuroimaging modality fusion in Alzheimer's classification using convolutional neural networks

Area of Science:

• Neurology
• Medical Imaging
• Artificial Intelligence

Background:

• Automated Alzheimer's disease (AD) classification holds significant clinical potential.
• Deep neural networks demonstrate high efficacy in AD classification using neuroimaging data.
• A comprehensive comparison between MRI and amyloid PET modalities is lacking.

Purpose of the Study:

• To compare the efficacy of MRI and amyloid PET imaging for Alzheimer's dementia classification.
• To evaluate the benefits of a multimodal fusion approach using both MRI and amyloid PET.
• To explore future applications of these imaging modalities in deep learning-based AD studies.

Main Methods:

• Utilized the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset.
• Employed identical deep neural network architectures for both modalities.
• Performed a comparative analysis of MRI and amyloid PET performance.
• Investigated a data fusion strategy combining MRI and amyloid PET.

Main Results:

• Both MRI and amyloid PET demonstrated effectiveness in Alzheimer's dementia classification.
• Analysis revealed the relative strengths of each imaging modality.
• The study explored the synergistic benefits of combining MRI and amyloid PET data.

Conclusions:

• Deep learning models can effectively classify Alzheimer's dementia using MRI and amyloid PET.
• Comparing individual modalities and their fusion provides valuable insights for AD diagnosis.
• Future AD research can leverage multimodal imaging data with deep learning for enhanced classification.