Accuracy of left ventricular volumes and ejection fraction measurements by contemporary three-dimensional echocardiography with semi- and fully automated software: systematic review and meta-analysis of 1,881 subjects

Background: Although left ventricular (LV) quantification software for transthoracic three-dimensional echocardiography (3DE) have been steadily advancing, the measurement accuracy of semiautomated and fully automated software has not been systematically investigated.

Methods: We performed a systematic review and meta-analysis following a search of the PubMed, Embase, and Scopus databases for studies that directly compared LV volumes and ejection fraction (EF) using 3DE with ultrasound machines manufactured by one of the four manufacturers with either semiautomated (n = 4) or fully automated software (n = 2) and cardiac magnetic resonance (CMR) during the period from January 1, 2011, to July 23, 2018. A random effects model was used to determine the bias and 95% CI of LV end-diastolic volume (EDV), end-systolic volume (ESV), and EF. A metaregression and subgroup analysis were performed to investigate intersoftware variabilities.

Results: Of a total of 38 studies (1,881 subjects), the pooled bias and 95% CI for EDV, ESV, and EF of semiautomated software were -39.3 mL (-49.2 to -29.5 mL, P < .001), -19.6 mL (-25.9 to -13.3 mL, P < .001), and -0.6% (-2.0% to 0.7%, P = .360), respectively. The corresponding values for fully automated software were -14.5 mL (-25.7 to -3.3 mL, P = .011), -6.3 mL (-11.4 to -1.2 mL, P = .016), and -1.1% (-3.5% to 1.3%, P = .356) compared with CMR. The EDV and ESV biases for 3DE and CMR became significantly smaller and less heterogeneous when fully automated software was used. A metaregression analysis revealed that EDV bias became larger with an increase in EDV when semiautomated software was used but not when fully automated software was used.

Conclusions: Although 3DE still underestimates LV volumes, EF showed similar values to CMR for both types of software. The adoption of fully automated software could further improve measurement accuracy with fewer observer variabilities.