Influence of phase correction of late gadolinium enhancement images on scar signal quantification in patients with ischemic and non-ischemic cardiomyopathy

Myocardial fibrosis imaging using late gadolinium enhancement (LGE) cardiac magnetic resonance (CMR) has been validated as a quantitative predictive marker for response to medical, surgical, and device therapy. To date, all such studies have examined conventional, non-phase corrected magnitude images.  However, contemporary practice has rapdily adopted phase-corrected image reconstruction. We sought to investigate the existence of any systematic bias between threshold-based scar quantification performed on conventional magnitude inversion recovery (MIR) and matched phase sensitive inversion recovery (PSIR) images.

In 80 patients with confirmed ischemic (N = 40), or non-ischemic (n = 40) myocardial fibrosis, and also in a healthy control cohort (N = 40) without fibrosis, myocardial late enhancement was quantified using a Signal Threshold Versus Reference Myocardium technique (STRM) at ≥2, ≥3, and ≥5 SD threshold, and also using the Full Width at Half Maximal (FWHM) technique. This was performed on both MIR and PSIR images and values compared using linear regression and Bland-Altman analyses.

Linear regression analysis demonstrated excellent correlation for scar volumes between MIR and PSIR images at all three STRM signal thresholds for the ischemic (N = 40, r = 0.96, 0.95, 0.88 at 2, 3, and 5 SD, p < 0.0001 for all regressions), and non ischemic (N = 40, r = 0.86, 0.89, 0.90 at 2, 3, and 5 SD, p < 0.0001 for all regressions) cohorts. FWHM analysis demonstrated good correlation in the ischemic population (N = 40, r = 0.83, p < 0.0001). Bland-Altman analysis demonstrated a systematic bias with MIR images showing higher values than PSIR for ischemic (3.3 %, 3.9 % and 4.9 % at 2, 3, and 5 SD, respectively), and non-ischemic (9.7 %, 7.4 % and 4.1 % at ≥2, ≥3, and ≥5 SD thresholds, respectively) cohorts. Background myocardial signal measured in the control population demonstrated a similar bias of 4.4 %, 2.6 % and 0.7 % of the LV volume at 2, 3 and 5 SD thresholds, respectively. The bias observed using FWHM analysis was −6.9 %.

Scar quantification using phase corrected (PSIR) images achieves values highly correlated to those obtained on non-corrected (MIR) images. However, a systematic bias exists that appears exaggerated in non-ischemic cohorts. Such bias should be considered when comparing or translating knowledge between MIR- and PSIR-based imaging.