1357790 - Comparisons between circumferential strains derived by feature tracking algorithms and myocardial tagging in childhood cancer survivors

Childhood cancer survivors with a history of anthracycline therapy (≥13 years of age, ≥2 years post-therapy) were prospectively enrolled. CMR imaging was performed, including steady-state free precession (SSFP) cine imaging in the left ventricular (LV) long axis and contiguous slices from the base to apex (7 mm thick, 30 phases). Myocardial tagged cines were acquired in the short axis orientation at the mid-LV level (7 mm thick, tag space 7 mm). Endocardial and epicardial borders were manually contoured on tagged images and analyzed using Segment (Medviso, Sweden). On SSFP Cine imaging, automated endocardial and epicardial contours were generated and manually adjusted using 3 FT methods: 2D and 3D algorithms using CVI42 (Circle Cardiovascular Imaging, Canada) and Segment. Two observers (AM, HB) performed each measurement and generated mid-LV average and segmental peak circumferential strain (CS). Interobserver reproducibility was assessed via coefficients of variation (CVs) and comparisons with tagged imaging were performed by describing the mean difference and Pearson’s correlation.

Results: CMR imaging studies performed on 13 participants were analyzed (Table 1). Interobserver CVs for average mid-LV CS was 7.7%, 6.7%, 9.4%, and 10.1% for Circle 2D, Circle 3D, Segment FT, and tagged image analysis, respectively (Table 2). Segmental CVs ranged from 6 to 28%; the lowest CVs for segmental strains were in the tagged analysis (average 11.9% across segments). For average CS, the mean bias between FT methods and tagged images was -1.6%, -4.1%, and -2.3% for Circle 2D, Circle 3D, and Segment, respectively, with FT methods producing more negative values (Table 3). Correlation between FT and tagged analysis was highest for the Circle 2D method and was generally poorer and less consistent for segmental measures.

Conclusion: Average strain values were generally reproducible across methods. Segmental reproducibility was worse overall but highest with tagged analysis. Average strain values derived by FT algorithms were correlated but with a small bias toward more negative values in comparison with tagged analysis, while segmental correlations were inconsistent. Although adequate, tagged analysis reproducibility was not as robust as expected; it is possible that a smaller tag space may improve reproducibility in childhood cancer survivors, given their predisposition to myocardial wall thinning. Further study in larger cohorts with more functional variation may provide additional insight into the use of these measures.