Quantitative R2∗ Map Synthesis from T1W and T2W Images Using Generative Adversarial Networks for Parkinson's Disease

Motivation: R2* mapping is important for measuring iron content in Parkinson's disease. However, conventional methods for obtaining R2* maps, involving multi-echo-gradient-recalled-echo sequences acquisition and image reconstruction, are both time-consuming and computationally intensive. Goal(s): Our goal is to synthesize accurate R2* maps quickly using a Generative Adversarial Network (GAN). Approach: A GAN model was developed to synthesize R2* maps from T1W and T2W images. The performance of synthetic R2* maps was evaluated using both internal and external datasets. Results: The synthetic R2* maps showed excellent performance in diagnosing and assessing Parkinson's disease across all test datasets, demonstrating the feasibility of our proposed GAN model. Impact: The generated R2* maps show good correlations with real R2* maps and have the potential to provide valuable information about the disease process. Our study provides a promising start for PD diagnosis, assessment and monitoring using synthetic R2* maps.