Generative Adversarial Networks Based on Fine-Grained Image Recognition for the Progression Prediction of Progressive...

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Neurology

Background:

• Progressive mild cognitive impairment (pMCI) can advance to Alzheimer's disease (AD), making early identification crucial for timely intervention.
• Current methods using multimodal neuroimaging data (MRI, PET) for pMCI prediction are limited by data scarcity and suboptimal cross-modal synthesis.
• Existing generative adversarial networks (GANs) often fail to capture local details in synthesized neuroimages, impacting prediction performance.

Purpose of the Study:

• To develop a novel generative adversarial network (FGGAN) for high-quality cross-modal neuroimage synthesis.
• To enhance the prediction accuracy of progressive mild cognitive impairment (pMCI) using fused multimodal neuroimaging features.
• To address the limitations of existing GANs in capturing local information for improved image synthesis.

Main Methods:

• Proposed a fine-grained generative adversarial network (FGGAN) incorporating a GAN, feature depth extraction (FDE) module, and a classifier.
• The GAN synthesizes missing neuroimaging modalities using local and global cues, extracting multimodal feature representations.
• The FDE module refines semantic features for improved classifier adaptation, enabling pMCI progression prediction from fused multimodal data.

Main Results:

• FGGAN demonstrated superior performance in synthesizing high-quality cross-modal neuroimages compared to existing methods.
• The proposed method achieved enhanced accuracy in classifying progressive mild cognitive impairment (pMCI) progression.
• Validation on the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset confirmed the effectiveness of FGGAN.

Conclusions:

• FGGAN effectively synthesizes multimodal neuroimaging data, overcoming limitations of previous approaches.
• The enhanced image synthesis quality directly contributes to improved prediction of pMCI progression.
• This approach holds significant clinical value for the early identification of individuals at risk of developing Alzheimer's disease.