Morphological Feature Visualization of Alzheimer's Disease via Multidirectional Perception GAN

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neuroscience

Background:

• Early diagnosis of Alzheimer's disease (AD) is crucial for effective treatment and slowing disease progression.
• Visualizing subtle morphological changes in early-stage AD is clinically significant for patient management.

Purpose of the Study:

• To propose a novel multidirectional perception generative adversarial network (MP-GAN) for visualizing Alzheimer's disease (AD) severity.
• To capture salient global features and delineate subtle lesions in MR images for different stages of AD.

Main Methods:

• Developed a MP-GAN with a multidirectional mapping mechanism for efficient global feature capture.
• Utilized class discriminative maps for MR image transformations between source and target domains.
• Integrated adversarial loss, classification loss, cycle consistency loss, and L1 penalty for multi-class learning.

Main Results:

• MP-GAN demonstrated superior performance on the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset compared to existing methods.
• The model successfully visualized subtle lesions consistent with clinical observations.
• Achieved clear delineation of subtle lesions through MR image transformations.

Conclusions:

• MP-GAN effectively visualizes morphological features indicating Alzheimer's disease severity.
• The proposed model aids in the early diagnosis and understanding of AD progression.
• This AI-driven approach shows promise for clinical application in neurodegenerative disease assessment.