Myocardial strains can be calculated using cardiovascular magnetic resonance (CMR) feature-tracking (FT) algorithms. They show excellent intra- and inter-observer agreement but rather disappointing inter-vendor agreement. Currently, it is unknown how well CMR-FT-based strain values agree with manually obtained strain values.
In 45 subjects (15 controls, 15 acute myocardial infarction, 15 non-ischemic dilated cardiomyopathy), end-systolic manually derived strains were compared to four CMR-FT software packages. Global radial strain (GRS), global circumferential strain (GCS) and global longitudinal strain (GLS) were determined. Intra- and inter-observer agreement and agreement between manual and CMR-FT analysis were calculated. Statistical analysis included Bland-Altman plots, intra-class correlation coefficient (ICC) and coefficient of variation (CV).
Manual contouring yielded excellent intra-observer (ICC 0.903 (GRS) to 0.995 (GCS)) and inter-observer agreement (ICC 0.915 (GRS) to 0.966 (GCS)) with CV ranging 4.7% (GCS) to 20.7% (GRS) and 12.7% (GCS) to 20.0% (GRS), for intra-observer and inter-observer agreement, respectively. Agreement between manual and CMR-FT strain values ranged from poor to excellent, with best agreement for GCS (ICC 0.857-0.935) and intermediate for GLS (ICC 0.591-0.914), while ICC values for GRS ranged widely (ICC 0.271-0.851). In particular, two software packages showed a strong trend toward systematic underestimation of myocardial strain in radial and longitudinal direction, correlating poorly to moderately with manual contouring, i.e., GRS (ICC 0.271, CV 25.2%) and GLS (ICC 0.591, CV 17.6%).
Some CMR-FT values agree poorly with manually derived strains, emphasizing to be cautious to use these software packages in the clinical setting. In particular, radial and longitudinal strain tends to be underestimated when using manually derived strains as reference.

Author