Accuracy and reproducibility of geometric models for assessment of left ventricular ejection faction using cardiovascular magnetic resonance

Abstract

Study objective: To assess the accuracy and reproducibility of geometric models for assessment of left ventricular ejection fraction. Background: Cardiovascular magnetic resonance (CMR) is considered the clinical gold standard for accurate and reproducible assessment of left ventricular (LV) ejection fraction (EF). However, manual contouring of an entire LV short-axis stack can be time consuming. A number of geometric approximation models for assessment of ventricular volumes have previously been validated using angiography, echocardiography and single photon emission computed tomography. We aimed to assess the accuracy and reproducibility of these geometric models for LV EF assessment in unselected patients referred for clinically indicated CMR. Methods: 67 consecutive patients were recruited. SSFP cine images were obtained using a 1.5 T scanner (Siemens Avanto, Germany) equipped with a 32-channel surface coil. LV volumetric analysis was performed with the open source software package OSIRIX utilising the following geometric models - Triplane, (Table Presented) Biplane, Monoplane, Hemi-ellipse, Modified Simpson's ellipse (“Mod Simps”) and Teichholz. Analysis by manual endocardial border tracing of each short-axis slice in an LV 'stack' using Siemens Argus software was used as the reference standard. The images of 25 randomly selected patients were also independently analysed by a second observer to allow assessment of interobserver reproducibility and then reanalysed by both observers to assess intra-observer reproducibility. Results: The EF obtained by every geometric model was significantly different to the EF obtained by the reference standard with wide Bland-Altman levels of agreement (Table 1). The inter-observer and intra-observer reproducibility for each model was low, also with wide Bland-Altman ranges (Table 2). Conclusion: The accuracy and reproducibility of geometric models for LV EF assessment are too low for clinical use.