Deep Learning Reconstruction Improved LGE Image Quality for Myocardial Scar Detection in Cardiomyopathy Patients with Ventricular Arrhythmias

Motivation: LGE is routinely used for detecting myocardial scar in cardiomyopathy patients. However, accurate detection remains challenging in ventricular arrhythmias (VAS) patients. Goal(s): Assess the impact of deep learning reconstruction (DLRecon) on LGE image quality for detecting myocardial scars in VAS. Approach: 27 VAS and 37 non-VAS cardiomyopathy patients were finally involved. LGE was imaged and separately reconstructed using conventional reconstruction (ConRecon) and DLRecon. Results: DLRecon can significantly improve the image quality of LGE for better presentation of scar-blood and scar-myocardium boundary in VAS. By using 5SD as threshold, DLRecon LGE showed significantly higher ratio of sudden cardiac death risk in VAS patients. Impact: DLRecon can improve the image quality of LGE for better delineating characteristics of lesion. Our results suggested the application of DLRecon based LGE might be beneficial for evaluating sudden cardiac death risk stratification in cardiomyopathy patients with VAS.