Tensorizing GAN With High-Order Pooling for Alzheimer's Disease Assessment

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Medical Diagnostics

Background:

• Early diagnosis of Alzheimer's disease (AD) and mild cognitive impairment (MCI) is crucial for effective treatment.
• Deep learning offers promising avenues for analyzing complex neuroimaging data.
• Magnetic Resonance Imaging (MRI) provides valuable structural information about the brain.

Purpose of the Study:

• To propose a novel deep learning model for the early diagnosis of AD and MCI.
• To leverage tensorization and high-order pooling for enhanced classification of neuroimaging data.
• To introduce the Tensor-train, High-order pooling and Semisupervised learning-based GAN (THS-GAN) as a pioneering approach for AD diagnosis using MR images.

Main Methods:

• Development of a novel tensorizing Generative Adversarial Network (GAN) incorporating a three-player cooperative game framework.
• Integration of a high-order pooling scheme to utilize second-order statistics from holistic MRI data.
• Application of semisupervised learning techniques for improved model training and sample generation.

Main Results:

• The proposed THS-GAN achieved superior classification performance on the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset compared to existing methods.
• Both tensor-train and high-order pooling components were shown to significantly enhance classification accuracy.
• Generated samples demonstrated the model's capability for effective semisupervised learning in AD diagnosis.

Conclusions:

• The THS-GAN model represents a significant advancement in the early diagnosis of AD and MCI using deep learning on MRI.
• The tensorization and high-order pooling strategies are effective in improving the analysis of neuroimaging data for disease classification.
• The developed model shows potential for clinical application in aiding early detection of neurodegenerative diseases.