Toward Replacing Late Gadolinium Enhancement With Artificial Intelligence Virtual Native Enhancement for Gadolinium-Free Cardiovascular Magnetic Resonance Tissue Characterization in Hypertrophic Cardiomyopathy

Abstract

Background: Late gadolinium enhancement (LGE) cardiovascular magnetic resonance (CMR) imaging is the gold standard for noninvasive myocardial tissue characterization but requires intravenous contrast agent administration. It is highly desired to develop a contrast agent-free technology to replace LGE for faster and cheaper CMR scans.

Methods: A CMR virtual native enhancement (VNE) imaging technology was developed using artificial intelligence. The deep learning model for generating VNE uses multiple streams of convolutional neural networks to exploit and enhance the existing signals in native T1 maps (pixel-wise maps of tissue T1 relaxation times) and cine imaging of cardiac structure and function, presenting them as LGE-equivalent images. The VNE generator was trained using generative adversarial networks. This technology was first developed on CMR datasets from the multicenter Hypertrophic Cardiomyopathy Registry, using hypertrophic cardiomyopathy as an exemplar. The datasets were randomized into 2 independent groups for deep learning training and testing. The test data of VNE and LGE were scored and contoured by experienced human operators to assess image quality, visuospatial agreement, and myocardial lesion burden quantification. Image quality was compared using a nonparametric Wilcoxon test. Intra- and interobserver agreement was analyzed using intraclass correlation coefficients (ICC). Lesion quantification by VNE and LGE were compared using linear regression and ICC.

Results: A total of 1348 hypertrophic cardiomyopathy patients provided 4093 triplets of matched T1 maps, cines, and LGE datasets. After randomization and data quality control, 2695 datasets were used for VNE method development and 345 were used for independent testing. VNE had significantly better image quality than LGE, as assessed by 4 operators (n=345 datasets; P<0.001 [Wilcoxon test]). VNE revealed lesions characteristic of hypertrophic cardiomyopathy in high visuospatial agreement with LGE. In 121 patients (n=326 datasets), VNE correlated with LGE in detecting and quantifying both hyperintensity myocardial lesions (r=0.77-0.79; ICC=0.77-0.87; P<0.001) and intermediate-intensity lesions (r=0.70-0.76; ICC=0.82-0.85; P<0.001). The native CMR images (cine plus T1 map) required for VNE can be acquired within 15 minutes and producing a VNE image takes less than 1 second.

Conclusions: VNE is a new CMR technology that resembles conventional LGE but without the need for contrast administration. VNE achieved high agreement with LGE in the distribution and quantification of lesions, with significantly better image quality.

Keywords: artificial intelligence; cardiomyopathy, hypertrophic; contrast media; deep learning; gadolinium; magnetic resonance imaging.

Figure 1.

Overview of the VNE imaging technology.A, Simplified illustration of Hypertrophic Cardiomyopathy Registry scan protocol which includes native (precontrast) cine, T1 mapping (including native inversion recovery–weighted images), and conventional postcontrast late gadolinium enhancement. B, VNE generator. Native cardiovascular magnetic resonance images are input to 3 steams of encoder–decoder U-nets to extract feature maps, followed by a further neural network block to fuse all feature maps and derive a VNE image. Once trained, producing a VNE image takes <1 s. VNE indicates virtual native enhancement.

Figure 2.

Flow of patient material selection for developing and testing the virtual native enhancement technology. *The excluded T1 maps (n=10) in testing materials are disclosed in Figure III in the Data Supplement. **Four of the 36 triplets were retrospectively excluded from analysis because of slice position mismatch and coil problems identified by consensus of 2 cardiovascular magnetic resonance experts (see Figure IV in the Data Supplement). These examples were not detected automatically using the predefined criteria in slice position matching, and were excluded after manual inspection. HCMR indicates Hypertrophic Cardiomyopathy Registry; LGE, late gadolinium enhancement; and VNE, virtual native enhancement.

Figure 3.

VNE and LGE image quality assessment on 346 test materials (124 patients).A, VNE provides significantly better image quality, as assessed by 4 blinded operators and their average scores (all P<0.001). For cases with “uninterpretable” (red clusters) or “poor” (blue) LGE images, VNE provides superior imaging quality in all but 1 case (dashed line). B, Examples of image quality improvement by VNE, which has more consistent appearance and defined borders. Arrows point to the LGE artefacts. LGE indicates late gadolinium enhancement; and VNE, virtual native enhancement.

Figure 4.

Examples to illustrate visuospatial agreement between VNE and conventional LGE. T1 colormaps (top row) were adjusted individually to highlight the T1 signals corresponding to VNE signals. The bottom 2 rows visualize lesion regions by VNE and LGE using progressive thresholding (full width at half, a quarter, and eighth maximum, ie, at 50th, 25th, and 12.5th percentiles) displayed with different colors. A through F, High visuospatial agreement was observed between VNE and LGE. Yellow arrows point to slightly different right ventricle sizes in VNE and LGE, suggesting patient movement between acquisitions. G, An example of VNE displaying subtle changes clearer than LGE. LGE indicates late gadolinium enhancement; and VNE, virtual native enhancement.

Figure 5.

VNE correlated strongly with conventional LGE in quantifying hyperintensity to intermediate-intensity lesions (left to right) in 121 test patients.A through C use the same thresholding methods FWHM, FWQM, and FWEM (ie, thresholding at 50th, 25th, and 12.5th percentiles, reflecting hyperintensity to intermediate-intensity subtle lesions) for VNE and LGE. D through F use adjusted thresholding at 35th, 20th, and 10th percentiles for VNE. Threshold values are illustrated on color bars. Linear regression equations, correlation coefficient R values, and ICCs are provided. Bland–Altman plots demonstrate perceivable trends (arrowed) with associated clustering, suggesting enhanced signals in VNE for subtle lesions. FWEM, full width at eighth maximum; FWHM, full width at half maximum; FWQM, full width at quarter maximum; ICC, intraclass correlation coefficient; LGE, late gadolinium enhancement; LV, left ventricle; and VNE, virtual native enhancement.