A comparison between methods for automatic quantification of global left ventricular function

Introduction: Volumetric cardiac ultrasound imaging has become more readily available in daily clinical practice. To date, quantitative analysis of these data sets requires a significant amount of user interaction. This not only results in significant post-processing times but also in a higher intra- and inter-observer variability. Recently, we introduced methods that could help in automating this process. On the one hand, an edge detection algorithm in combination with a deformable subdivision surface was presented for automatic segmentation of the LV cavity. A real-time, dynamic implementation (RCTL) of this segmentation approach in combination with a Kalman filter allows tracking the subendocardial boundary throughout the cardiac cycle. On the other hand, an automatic 3D motion estimation algorithm was presented in which subsequent image volumes are elastically registered using a B-spline transformation field (splineMIRIT). Aim: Functional parameters derived from volume changes that occur during the cardiac cycle, i.e. stroke volume (SV) and ejection fraction (EF), can thus be estimated using either RCTL or splineMIRIT in order to get a fully automated analysis of LV function. The aim of the present study was to contrast both methods for their accuracy in a clinical setting. Methods: Volumetric ultrasound data were acquired in 5 healthy volunteers and 8 patients with coronary artery disease at a frame rate of 28.1±3.9Hz using a GE Vivid7 Dimensions (GE Vingmed, Norway). The end-diastolic (ED) volume was automatically segmented using RCTL and the resulting subendocardial mesh was subsequently tracked using both RCTL and splineMIRIT. As a reference, both ED and end-systolic volumes were manually contoured using commercially available tools (EchoPac, GE Vingmed, Norway). From both automatically and manually defined volumes, SV and EF were derived. Correlations of the automatic methods against the reference method were made as well as a Bland-Altman analysis to determine agreement. Results: The correlations and Bland-Altman plots for the EF are shown in Figure 1. The regression coefficients (0.88 and 1.03) were not statistically significantly different (p = 0.66) with similar results for SV (R=0.81 and R=0.86). Bias and limits of agreement were -6.5 [-17.2 - +4.2]% and -9.6 [-19.0 - +0.3]% respectively. While there was no systematic difference between splineMIRIT and RCTL for the normal volunteers, splineMIRIT gave lower values (38±7%) than RCTL (42±7%) for the infarct patients (p