Feasibility of two-dimensional multi-segmented late gadolinium enhancement combined with artificial intelligence reconstruction deep-learning noise reduction in patients with non-ischemic cardiomyopathy

Although two-dimensional (2D) single-segmented late gadolinium enhancement (2D-SSLGE) sequences are the gold standard for LGE acquisition, 2D multi-segmented LGE (2D-MSLGE) and artificial intelligence reconstruction deep-learning noise reduction (NR) show a potential to improve image quality and reduce scan times. Thus, this study aimed to assess the feasibility of 2D-MSLGE reconstructed with NR in patients with non-ischemic cardiomyopathy (NICM).

This study prospectively enrolled 60 patients (39 men, mean age, 57.7 ± 16.5 years) with NICM. Subjective overall image quality was evaluated using a 5-point Likert scale. Artifact visibility was also scored. LGE mass was manually quantified using a semi-automated grayscale threshold with a cutoff value of 5 standard deviations.

The median actual acquisition time of 2D-MSLGE reconstructed with NR (123 s; range, 81-246) was significantly shorter than that of 2D-SSLGE reconstructed with NR (540 s; range, 420-720 s, p < 0.001). There was no significant difference in overall image quality between 2D-MSLGE reconstructed with NR and 2D-SSLGE reconstructed with NR (median, 3.0; IQR, 3.0-3.0 vs. 3.0; IQR, 3.0-4.0, p = 0.197). Image noise was significantly reduced in 2D-MSLGE reconstructed with NR (3.0; 3.0-4.0) compared with that in 2D-SSLGE reconstructed with NR (3.0; 2.0-4.0, p = 0.026). No significant difference was found in the LGE mass between 2D-MSLGE reconstructed with NR (16.9%; 10.9-25.1) and 2D-SSLGE reconstructed with NR (17.3%; 11.4-27.0, p = 0.939).

2D-MSLGE images reconstructed with NR could be acquired at shorter scan times and provide accurate LGE in patients with NICM.