Comprehensive Summary
This paper investigates the usage of a novel Virtual Native Enhancement (VNE) cardiovascular magnetic resonance imaging technology to identify myocardial fibrosis and tissue scarring. The AI model was compared to the traditional diagnostic tool in MRI-Imaging, Late Gadolinium Enhancement (LGE). The VNE model consists of convolutional neural networks that analyze native T1 maps and cine imaging which are predictors of early myocardial changes in patients with hypertrophic cardiomyopathy (HCM). The VNE images were found to have significantly better perceived image quality when compared to LGE, while maintaining a high agreement rate in regards to quantification of myocardial lesions (r=0.70-0.79, ICC=0.77-0.87, P<0.001). Furthermore, the VNE model was able to analyze the inputs within 15 minutes and generate images within one second, much faster than the traditional LGE process. The main takeaways from this study include that VNE serves as an accurate predictor of myocardial lesions, while being less invasive, faster, and cheaper than the standard usage of LGE as the main cardiovascular magnetic resonance imaging.
Outcomes and Implications
Currently, late gadolinium enhancement cardiovascular magnetic resonance imaging sets the gold standard for detection of myocardial fibrosing and scarring. However, this process requires an intravenous injection of a gadolinium-based contrast agent (GBCA). GBCA can be harmful for patients with kidney failure or those with GBCA allergies. The introduction of the VNE model offers an alternative diagnostic model that can shorten scan times, drastically reduce costs, and shorten test time, without requiring the exposure of repeat gadolinium in patients with HCM or other cardiomyopathies. While further validation and approval are needed before clinical implementation, these preliminary results are promising for the use of artificial intelligence in myocardial tissue characterization.