High-resolution conditional MR image synthesis through the PACGAN framework.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Deep Learning

Background:

• Deep learning for medical images faces challenges like limited data, overfitting, and imbalanced datasets.
• Synthetic datasets offer a solution by enabling control over size and balance.

Purpose of the Study:

• To present PACGAN (Progressive Auxiliary Classifier Generative Adversarial Network), a framework for generating high-quality, class-specific synthetic medical images.
• To address data limitations in deep learning for medical image analysis, specifically for Alzheimer's disease research.

Main Methods:

• Developed PACGAN, combining Progressive Growing GAN and Auxiliary Classifier GAN (ACGAN).
• Utilized latent space information for conditional synthesis of high-resolution brain MR images.
• Trained the framework on the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset.

Main Results:

• PACGAN generated realistic synthetic brain MR images for Alzheimer's disease patients and healthy controls.
• Quantitative metrics assessed the quality of the synthetic images.
• The pre-trained discriminator achieved an AUC of 0.813 in classifying real unseen images, indicating successful class characteristic capture.

Conclusions:

• PACGAN effectively generates high-quality, class-specific synthetic medical images.
• The framework shows promise in augmenting datasets for deep learning in medical imaging, particularly for Alzheimer's disease.
• Open-source code and pre-trained models are available for further research.