Generation of contrast-enhanced cardiac MRI from contrast-free scans: a multi-center, multi-manufacturer study.
Authors
Affiliations (10)
Affiliations (10)
- College of Bioengineering, Chongqing Emergency Medical Centre, Chongqing University Central Hospital, Chongqing University, 400014, Chongqing, China.
- Bio-Med Informatics Research Centre, Xinqiao Hospital, Army Medical University, 40037, Chongqing, China.
- Department of Radiology, The First Affiliated Hospital of Chongqing Medical University, 400016, Chongqing, China.
- Department of Radiology, Southwest Hospital, Third Military Medical University (Army Medical University), 40038, Chongqing, China.
- Department of Radiology, The Second Affiliated Hospital of Army Medical University, 40037, Chongqing, China.
- School of Medical Technology, Beijing Institute of Technology, 100081, Beijing, China.
- Department of Cardiovascular Surgery, Xinqiao Hospital, 40037, Chongqing, China.
- Center for Biomedical Imaging Research, School of Biomedical Engineering, Tsinghua University, 100084, Beijing, China.
- College of Bioengineering, Chongqing Emergency Medical Centre, Chongqing University Central Hospital, Chongqing University, 400014, Chongqing, China. [email protected].
- Bio-Med Informatics Research Centre, Xinqiao Hospital, Army Medical University, 40037, Chongqing, China. [email protected].
Abstract
Late gadolinium enhancement (LGE) cardiac magnetic resonance imaging (MRI) is regarded as the non-invasive gold standard for myocardial tissue characterization. However, its reliance on gadolinium-based contrast agents limits its applicability in patients with renal impairment or contrast allergies. We develop Cine2LGE, a method that simultaneously generates LGE-like images and estimates enhancement probability directly from contrast-free cine MRI. The model was trained and evaluated on a multi-center, multi-manufacturer, multi-disease dataset, and externally tested on an independent dataset. Image quality was assessed by three independent, experienced radiographers. Visualization consistency between LGE-like and real LGE images was evaluated through linear regression, correlation coefficient analysis, and Bland-Altman plots. The accuracy of enhancement probability prediction was assessed using the area under the receiver operating characteristic curve (AUC). The training dataset comprised 6361 paired cine and LGE images from 1189 subjects. The internal test dataset included 777 paired images from 129 subjects, and the external test dataset consisted of 1055 paired images from 198 subjects. Cine2LGE significantly outperformed conventional LGE in terms of image quality (p < 0.05). Strong correlations were observed between Cine2LGE and LGE in the quantification of both hyperintensity (Pearson correlation, 0.732/0.833; p < 0.05) and intermediate-intensity myocardial lesions (Pearson correlation, 0.759/0.712; p < 0.05). The enhancement probability prediction showed robust performance on both internal and external test sets, with AUC values of 0.906 and 0.895, respectively. Cine2LGE offers a viable alternative for patients who cannot undergo conventional LGE imaging, enhancing both the accessibility and clinical utility of cardiac MRI. Question Can contrast-free cine MRI generate LGE-like images and accurately detect myocardial enhancement for patients contraindicated to gadolinium-based contrast agents? Findings Cine2LGE, a diffusion model-based method, generated LGE-like images with superior quality to conventional LGE and achieved high accuracy in predicting enhancement probability from cine MRI. Clinical relevance Cine2LGE shows promise in providing a valuable, contrast-free alternative for patients with LGE contraindications, potentially expanding both the accessibility and clinical utility of cardiac MRI.