Variability and agreement in the assessment of left ventricular ejection fraction and volumens applying automated tracking by 2D and 3D echocardiographic techniques

Background: Machine learning and use of automated tracking has improved the variability in the measurement of left ventricular (LV) ejection fraction (EF) in both twodimensional (2D) and threedimensional (3D) echocardiography techniques. However, further studies are required to verify this.
Purpose: The objective of the current study was to compare the interobserver
variability and agreement in the measurement of EF and volumes by applying 2D and 3D methods using automated tracings.
Methods: A total of 150 patients, aquired from The Østerbro Study, in sinus rhythm standard 2D apical and 3D full volume mode over 6 consecutive beats in single breath hold were acquired. None of the patients had structural heart disease. All echocardiograms were acquired using GE Vivid E9 ultrasound system and offline analysis were performed using EchoPAC v. 201. EF was determined by automated tracking using biplane Simpsons rule (2D AutoEF) and automated tracking of 3D volumes (4D AutoLVQ). Results are reported as limits of agreement and bias using Bland Altman plot along with intra class correlation coefficient and coefficient of variability (CV).
Results: The mean EF by 2D AutoEF and 4D AutoLVQ techniques were 59%±5% and 58%±5%. Volumes were smaller by 2D in comparison to 3D measurements: endsystolic volumes (ESV) 39ml±12 vs. 45ml±14 and enddiastolic volumes (EDV) 99ml±26 vs 108ml±28. Among the two techniques 4D AutoLVQ had the best agreement and least variability for measurements of both LV volume and EF (Table). The variability between the two EF estimation methods was limited (Bias 0.2±9%, CV 5%) (Table). Conclusion: Interobserver agreement was acceptable in the measurement of LV volumes and EF applying 2D and 3D automated techniques and can be applied in routine clinical practice.