RESUMEN
OBJECTIVES: Dark-blood late gadolinium enhancement (DB-LGE) cardiac magnetic resonance has been proposed as an alternative to standard white-blood LGE (WB-LGE) imaging protocols to enhance scar-to-blood contrast without compromising scar-to-myocardium contrast. In practice, both DB and WB contrasts may have clinical utility, but acquiring both has the drawback of additional acquisition time. The aim of this study was to develop and evaluate a deep learning method to generate synthetic WB-LGE images from DB-LGE, allowing the assessment of both contrasts without additional scan time. MATERIALS AND METHODS: DB-LGE and WB-LGE data from 215 patients were used to train 2 types of unpaired image-to-image translation deep learning models, cycle-consistent generative adversarial network (CycleGAN) and contrastive unpaired translation, with 5 different loss function hyperparameter settings each. Initially, the best hyperparameter setting was determined for each model type based on the Fréchet inception distance and the visual assessment of expert readers. Then, the CycleGAN and contrastive unpaired translation models with the optimal hyperparameters were directly compared. Finally, with the best model chosen, the quantification of scar based on the synthetic WB-LGE images was compared with the truly acquired WB-LGE. RESULTS: The CycleGAN architecture for unpaired image-to-image translation was found to provide the most realistic synthetic WB-LGE images from DB-LGE images. The results showed that it was difficult for visual readers to distinguish if an image was true or synthetic (55% correctly classified). In addition, scar burden quantification with the synthetic data was highly correlated with the analysis of the truly acquired images. Bland-Altman analysis found a mean bias in percentage scar burden between the quantification of the real WB and synthetic white-blood images of 0.44% with limits of agreement from -10.85% to 11.74%. The mean image quality of the real WB images (3.53/5) was scored higher than the synthetic white-blood images (3.03), P = 0.009. CONCLUSIONS: This study proposed a CycleGAN model to generate synthetic WB-LGE from DB-LGE images to allow assessment of both image contrasts without additional scan time. This work represents a clinically focused assessment of synthetic medical images generated by artificial intelligence, a topic with significant potential for a multitude of applications. However, further evaluation is warranted before clinical adoption.
RESUMEN
Dark-blood late gadolinium enhancement (LGE) has been shown to improve the visualization and quantification of areas of ischemic scar compared to standard bright-blood LGE. Recently, the performance of various semi-automated quantification methods has been evaluated for the assessment of infarct size using both dark-blood LGE and conventional bright-blood LGE with histopathology as a reference standard. However, the impact of this sequence on different quantification strategies in vivo remains uncertain. In this study, various semi-automated scar quantification methods were evaluated for a range of different ischemic and non-ischemic pathologies encountered in clinical practice. A total of 62 patients referred for clinical cardiovascular magnetic resonance (CMR) were retrospectively included. All patients had a confirmed diagnosis of either ischemic heart disease (IHD; n = 21), dilated/non-ischemic cardiomyopathy (NICM; n = 21), or hypertrophic cardiomyopathy (HCM; n = 20) and underwent CMR on a 1.5 T scanner including both bright- and dark-blood LGE using a standard PSIR sequence. Both methods used identical sequence settings as per clinical protocol, apart from the inversion time parameter, which was set differently. All short-axis LGE images with scar were manually segmented for epicardial and endocardial borders. The extent of LGE was then measured visually by manual signal thresholding, and semi-automatically by signal thresholding using the standard deviation (SD) and the full width at half maximum (FWHM) methods. For all quantification methods in the IHD group, except the 6 SD method, dark-blood LGE detected significantly more enhancement compared to bright-blood LGE (p < 0.05 for all methods). For both bright-blood and dark-blood LGE, the 6 SD method correlated best with manual thresholding (16.9% vs. 17.1% and 20.1% vs. 20.4%, respectively). For the NICM group, no significant differences between LGE methods were found. For bright-blood LGE, the 5 SD method agreed best with manual thresholding (9.3% vs. 11.0%), while for dark-blood LGE the 4 SD method agreed best (12.6% vs. 11.5%). Similarly, for the HCM group no significant differences between LGE methods were found. For bright-blood LGE, the 6 SD method agreed best with manual thresholding (10.9% vs. 12.2%), while for dark-blood LGE the 5 SD method agreed best (13.2% vs. 11.5%). Semi-automated LGE quantification using dark-blood LGE images is feasible in both patients with ischemic and non-ischemic scar patterns. Given the advantage in detecting scar in patients with ischemic heart disease and no disadvantage in patients with non-ischemic scar, dark-blood LGE can be readily and widely adopted into clinical practice without compromising on quantification.