Artificial Intelligence for Contrast-Free MRI: Scar Assessment in Myocardial Infarction Using Deep Learning-Based Virtual Native Enhancement

Abstract

Background: Myocardial scars are assessed noninvasively using cardiovascular magnetic resonance late gadolinium enhancement (LGE) as an imaging gold standard. A contrast-free approach would provide many advantages, including a faster and cheaper scan without contrast-associated problems.

Methods: Virtual native enhancement (VNE) is a novel technology that can produce virtual LGE-like images without the need for contrast. VNE combines cine imaging and native T1 maps to produce LGE-like images using artificial intelligence. VNE was developed for patients with previous myocardial infarction from 4271 data sets (912 patients); each data set comprises slice position-matched cine, T1 maps, and LGE images. After quality control, 3002 data sets (775 patients) were used for development and 291 data sets (68 patients) for testing. The VNE generator was trained using generative adversarial networks, using 2 adversarial discriminators to improve the image quality. The left ventricle was contoured semiautomatically. Myocardial scar volume was quantified using the full width at half maximum method. Scar transmurality was measured using the centerline chord method and visualized on bull's-eye plots. Lesion quantification by VNE and LGE was compared using linear regression, Pearson correlation (R), and intraclass correlation coefficients. Proof-of-principle histopathologic comparison of VNE in a porcine model of myocardial infarction also was performed.

Results: VNE provided significantly better image quality than LGE on blinded analysis by 5 independent operators on 291 data sets (all P<0.001). VNE correlated strongly with LGE in quantifying scar size (R, 0.89; intraclass correlation coefficient, 0.94) and transmurality (R, 0.84; intraclass correlation coefficient, 0.90) in 66 patients (277 test data sets). Two cardiovascular magnetic resonance experts reviewed all test image slices and reported an overall accuracy of 84% for VNE in detecting scars when compared with LGE, with specificity of 100% and sensitivity of 77%. VNE also showed excellent visuospatial agreement with histopathology in 2 cases of a porcine model of myocardial infarction.

Conclusions: VNE demonstrated high agreement with LGE cardiovascular magnetic resonance for myocardial scar assessment in patients with previous myocardial infarction in visuospatial distribution and lesion quantification with superior image quality. VNE is a potentially transformative artificial intelligence-based technology with promise in reducing scan times and costs, increasing clinical throughput, and improving the accessibility of cardiovascular magnetic resonance in the near future.

Keywords: artificial intelligence; cicatrix; magnetic resonance imaging; myocardial infarction.

Figure 1.

Deep learning approach of VNE for myocardial infarction. A, The neural network that combines cine frames and T1 maps (including inversion recovery–weighted images) and produces virtual native enhancement (VNE) images. B, Training VNE generator with a modified conditional generative adversarial network (cGAN) approach. C, Simultaneous training of 2 discriminators D1 and D2. LGE indicates late gadolinium enhancement; and MRI, magnetic resonance imaging.

Figure 2.

Flow of patient selection for VNE development and testing using clinical data sets. Clinical data sets used were from the University of Oxford Centre for Clinical Magnetic Resonance Research (OCMR) and the OxAMI study (Oxford Acute Myocardial Infarction). The training data set underwent strict late gadolinium enhancement (LGE) quality control to train the neural network to produce good-quality virtual native enhancement (VNE) images. The test data set went through initial rejection followed by multiobserver quality control. Rejected test data are available in Figures S1 and S4. *The generative adversarial network (GAN) translating postcontrast cine is specified in Supplemental Material 3. GBCA indicates gadolinium-based contrast agent; and LGE, late gadolinium enhancement.

Figure 3.

Examples to illustrate high agreement between LGE and contrast agent–free VNE for the visuospatial distribution and transmurality of myocardial scar. A through D, Three short-axis slices of late gadolinium enhancement (LGE) and virtual native enhancement (VNE) images of the same patient are shown on the left (color masks were used to depict areas of scar as orange and noninfarcted myocardium as dark blue); on the right, scar transmurality measured by LGE and VNE is shown, suggesting the likelihood of myocardial viability (0 to 25%, viable; 26% to 50%, likely viable; 51% to 75%, likely nonviable; 76% to 100%, nonviable). Dashed lines delineate presumed boundaries between myocardial territories. Arrows point to the areas of scar. LAD indicates left anterior descending artery; LCx, left circumflex artery; and RCA, right coronary artery.

Figure 4.

Assessment of myocardial scar size and transmurality using VNE. Virtual native enhancement (VNE) correlated strongly with late gadolinium enhancement (LGE) in scar size as a volume fraction of the sampled left ventricular (LV) myocardium (A) and in quantifying the mean transmurality of scarred chords per patient (B) in 66 test patients. Mean scar transmurality for each patient was calculated by averaging the transmural scar extent across all the chords (using the centerline chord method) that had at least 1% of scar extent on all available slices. Top, Correlation plots with the linear regression equations, Pearson correlation coefficients (R), statistical significance of correlation (P value), and intraclass correlation coefficients (ICC; 95% CI shown in brackets) provided. Bottom, Bland-Altman plots to analyze any systematic differences between quantification by VNE and LGE.

Figure 5.

Histopathologic comparison of VNE on 2 porcine model cases 8 to 9 weeks after myocardial infarction. Infarction was induced with ligation of the left anterior descending artery for 90 minutes followed by reperfusion. In both cases, virtual native enhancement (VNE) detected chronic myocardial infarction (arrows) in the left anterior descending territory (A) and was in high visual agreement with late gadolinium enhancement (LGE; B) and the ex vivo pathologic slices. These slices demonstrate macroscopic evidence of infarction (C); the infarcted region is pale pink on hematoxylin & eosin (H&E) staining (D), with collagen accumulation shown in light blue on Masson trichrome stain (E).