Signal Thresholding Segmentation of Ventricular Volumes in Young Patients with Various Diseases-Can We Trust the Numbers?

Abstract

In many cardiac diseases, right and left ventricular volumes in systole and diastole are diagnostically and prognostically relevant. Measurements are made by segmentation of the myocardial borders on cardiac magnetic resonance (CMR) images. Automatic detection of myocardial contours is possible by signal thresholding techniques, but must be validated before use in clinical settings. Biventricular volumes were measured in end-diastole (EDVi) and in end-systole (ESVi) both manually and with the MassK application, with signal thresholds at 30%, 50%, and 70%. Stroke volumes (SV) and cardiac indices (CI) were calculated from volumetric measurements and from flow measured in the ascending aorta and the main pulmonary artery, and both methods were compared. Reproducibility of volumetric measurements was tested in 20 patients. Measurements were acquired in 94 patients aged 15 ± 9 years referred for various conditions. EDVi and ESVi of both ventricles were largest with manual segmentation and inversely proportional to the MassK threshold. Manual and k30 SV and CI corresponded best to flow measurements. Interobserver variability was low for all volumes manually and with MassK. In conclusion, manual and 30% threshold-based biventricular volume segmentation agree best with two-dimensional, phantom-corrected phase contrast flow measurements in a young cardiac referral population and are well reproducible.