Background:
Quantification of myocardial scar using late gadolinium enhancement (LGE) is essential for diagnostic and prognostic assessment in cardiac MRI, yet multiple threshold-based techniques are used in clinical practice without standardized guidelines. Threshold-based techniques such as 2SD, 3SD, 5SD, full width at half maximum (FWHM), and automated algorithms are widely used, but their agreement relative to clinical reference standards remains uncertain. The 5SD method is typically recommended for ischemic and 3SD for non-ischemic scars.
Methods:
In 272 participants of a population-based (mean age 67 years, 56% female) with LGE-positive scans, left ventricular myocardial scar mass was manually segmented on PSIR images and quantified using five threshold approaches.
The cohort was derived from a large, ongoing, prospective, population-based health study in northern Germany that comprehensively investigates cardiovascular, metabolic, and other chronic diseases using standardized imaging, laboratory testing, and detailed clinical phenotyping.
For ischemic scars, 5SD served as reference; for non-ischemic scars, 3SD. Group differences were assessed by paired testing, two-way ANOVA, and regression analyses evaluating systematic and mass-dependent bias.
Results:
Quantified scar burden differed significantly between LGE quantification methods and varied according to scar etiology. In ischemic scars, 5SD yielded the lowest scar mass (8.1 ± 8.0 g), whereas 2SD (24.8 ± 13.3 g), 3SD (16.0 ± 10.5 g), FWHM (16.4 ± 12.9 g), and Auto-thresholding (36.7 ± 12.9 g) produced significantly higher values (all p < 0.001). Bland–Altman analyses demonstrated proportional bias for 2SD (β = 0.29, p = 0.001) and 3SD (β = 0.17, p = 0.04), while FWHM (β = -0.51, p < 0.001) and Auto-thresholding (β = -0.90, p < 0.001) markedly overestimated smaller scars.
In non-ischemic scars, 3SD yielded intermediate scar masses (7.8 ± 5.6 g) and served as the reference method. Compared with 3SD, 2SD (16.2 ± 9.3 g), FWHM (28.5 ± 26.1 g), and Auto-thresholding (41.9 ± 16.3 g) significantly overestimated scar burden, whereas 5SD markedly underestimated scar mass (1.9 ± 2.0 g; all p < 0.001). Bland–Altman analyses confirmed substantial mass-dependent bias for all alternative methods, with increasing divergence for 2SD (β = 0.57, p < 0.001) and pronounced overestimation of small scars by 5SD (β = -0.71, p < 0.001), FWHM (β = -0.97, p = 0.001), and Auto-thresholding (β = -0.68, p < 0.001).
Overall, method-dependent differences were considerably more pronounced in non-ischemic than in ischemic scar patterns.
Conclusion:
LGE-derived scar burden is strongly influenced by the selected quantification technique. Different methods are not interchangeable and may lead to clinically relevant differences in measured scar size. These discrepancies are particularly relevant for non-ischemic scar patterns and for studies evaluating prognostic thresholds. Standardization of LGE quantification is required to improve comparability across studies and to facilitate the implementation of scar burden as a quantitative imaging biomarker in clinical risk stratification.
