Cardiovascular magnetic resonance to predict appropriate implantable cardioverter defibrillator therapy in ischemic and nonischemic cardiomyopathy patients using late gadolinium enhancement border zone comparison of four analysis methods

Background: Late gadolinium enhancement (LGE) border zone on cardiac magnetic resonance imaging has been proposed as an independent predictor of ventricular arrhythmias. The purpose was to determine whether size and heterogeneity of LGE predict appropriate implantable cardioverter defibrillator (ICD) therapy in ischemic cardiomyopathy (ICM) and nonischemic cardiomyopathy (NICM) patients and to evaluate 4 LGE border-zone algorithms. Methods and Results: ICM and NICM patients who underwent LGE cardiac magnetic resonance imaging prior to ICD implantation were retrospectively included. Two semiautomatic algorithms, expectation maximization, weighted intensity, a priori information and a weighted border zone algorithm, were compared with a modified full-width half-maximum and a 2-3SD threshold-based algorithm (2-3SD). Hazard ratios were calculated per 1% increase in LGE. A total of 74 ICM and 34 NICM were followed for 63 months (1-140) and 52 months (0-133), respectively. ICM patients had 27 appropriate ICD events, and NICM patients had 7 ICD events. In ICM patients with primary prophylactic ICD, LGE border zone predicted ICD therapy in univariable and multivariable analysis measured by the expectation maximization, weighted intensity, a priori information, weighted border zone, and modified full-width half-maximum algorithms (hazard ratios 1.23, 1.22, and 1.05, respectively; P<0.05; negative predictive value 92%). For NICM, total LGE by all 4 methods was the strongest predictor (hazard ratios, 1.03-1.04; P<0.05), though the number of events was small. Conclusions: Appropriate ICD therapy can be predicted in ICM patients with primary prevention ICD by quantifying the LGE border zone. In NICM patients, total LGE but not LGE border zone had predictive value for ICD therapy. However, the algorithms used affects the predictive value of these measures.