LGEDiffusion: A Multi-Sequence Guided Diffusion Model for Virtual Contrast-Free LGE Generation in Myocardial Infarction

Motivation: Virtual LGE generation technology could reduce cardiac MRI (CMR) scan time, avoid gadolinium-based contrast agent (GBCA) risks, and benefit patients with GBCA contraindications. Goal(s): To develop and evaluate a multi-sequence CMR-guided virtual LGE generation technique, focusing on accuracy, effectiveness, and stability. Approach: A virtual LGE generation model based on a diffusion model was developed, leveraging Cine motion and T2 edema information for condition guidance. Results: Virtual LGE demonstrated high consistency with native LGE for identifying myocardial infarction lesions in terms of location and size, suggesting potential clinical applicability. Impact: This study highlights the potential of denoising diffusion probabilistic models for multi-sequence-guided MRI translation, emphasizing the value of virtual LGE as a viable contrast-free imaging alternative for myocardial infarction assessment.