T1 and T2 Mapping Reconstruction Based on Conditional DDPM.

Yansong Li,Lulu Zhao,Yun Tian,Shifeng Zhao

DOI: https://doi.org/10.1007/978-3-031-52448-6_29

2024-01-01

Abstract:Cardiac magnetic resonance imaging (CMR) has emerged as a crucial imaging modality for the diagnosis of cardiac diseases. T1 and T2 mapping are essential techniques for detecting cardiomyopathies. However, the imaging speed is noticeably slow and conventional mapping models often struggle to produce accurate results when the imaging process is compromised. To overcome this limitation, accelerated mapping techniques have been developed to reduce motion artifacts and enhance image quality. In this study, we propose a novel reconstruction method based on a conditional denoising diffusion probabilistic model (CDDPM). By utilizing accelerated mapping as a conditioning factor and iteratively applying a denoising process, we generate refined T1 and T2 maps from initially corrupted data consisting of pure Gaussian noise. The experimental results of the CMR Reconstruction Challenge demonstrate the effectiveness of our proposed method. Objective indicators show significant improvements, indicating enhanced image quality. Furthermore, our method successfully improves the texture quality of the images, providing more detailed and accurate information for cardiomyopathy diagnosis.