Improved reproducibility of right ventricular volumes and function estimation from cardiac magnetic resonance images using level-set models

Abstract

This study aims to assess whether an alternative method, that is based on volumetric surface detection (VoSD) without tracing and is totally free of geometric assumptions, can improve the reproducibility of right ventricular (RV) volume quantification from cardiac magnetic resonance (CMR) images, in comparison with a conventional disk-area technique. In a sample of 23 patients, with wide variability of RV end-diastolic volume (EDV: 47-131 ml), end-systolic volume (ESV: 20-76 ml), and ejection fraction (EF: 29-73%), using the standard method (Argus, Siemens) as the reference, the VoSD method showed good agreement for EDV, ESV, and EF estimations (correlation coefficient: 0.91, 0.94, and 0.94; Bland-Altman biases: 1 ml, 1 ml, and 0%; limits of agreement: +/-16 ml, +/-11 ml, and +/-11%, respectively). An analysis of the reproducibility of the two methods showed lower intraobserver variability for the VoSD method than for the conventional method, as evidenced by the coefficient of variability (CoV) values (2-6% vs. 8-15%; P < 0.05). In addition, the VoSD method showed improved interobserver reproducibility (7-10% vs. 8-15%), but the difference was statistically significant only for EF estimation variability (8 vs. 15%, P < 0.05). In conclusion, the newly developed VoSD technique allows accurate measurements of RV volumes and function, and appears to be more reproducible than the conventional methodology.