Automated reading shows good agreement with manual core-lab reading in high-resolution and in handheld transthoracic echocardiograms – Results from the PAVE project

Background: We recently showed that population data-based machine-learning can improve the automated quantification of cardiac structure and function in high-resolution transthoracic echocardiography (TTE).

Purpose: To explore the performance of automated reading in lower resolution echocardiograms acquired with a handheld device.

Methods and Results: PAVE (Pathology-oriented reading of echocardiography) is a cooperation project between University Hospital Wuerzburg (UKW) and TOMTEC Imaging Systems using an established federated machine-learning environment. We recruited 2043 patients (64±16 years, 44% women) presenting at the UKW for TTE. After high-resolution standard TTE (Vivid E95, GE) trained sonographers acquired images with a handheld ultrasound device (Lumify 2.0, Philips) according to a pre-specified protocol. Images of both modalities were loaded into the analysis platform of the Academic CoreLab Ultrasound-based Cardiovascular Imaging (Ultrasound Workspace®, TOMTEC, Germany). For the current analysis, we selected n=51 random patients (65±16 years, 39% women) from the PAVE cohort. Reading of high-resolution and handheld TTE was performed by a trained sonographer (>14 days apart and blinded to the reading results) as well as by the automated detector.

We present first results regarding 2D (parasternal long axis) measurements of interventricular septum (IVS), left ventricular diameter (LVD), posterior wall thickness (PW) at end-diastole and left atrial volume (LAvol) in apical two- (A2C) and four-chamber views (A4C) from manual CoreLab reading (CL) and automated detector reading (AD) of high-resolution TTE (mean±SD: IVS 11.3±4.1 mm, 10.9±2.5 mm; LVD 48.3±8.1 mm, 49.5±7.4 mm; PW 9.3±3.2 mm, 10.0±1.7 mm; LAvol-A2C 72.1±36.2 cm³, 79.5±37.9 cm³; LAvol-A4C 69.3±29.2 cm³, 78.4±37.2 cm³) as well as of the respective handheld echocardiograms (IVS 10.7±3.4 mm, 11.0±2.2 mm; LVD 47.7±7.4 mm, 48.3±6.6 mm; PW 9.7±2.7 mm, 10.2±1.6 mm; LAvol-A2C 72.9±31.8 cm³, 77.7±32.6 cm³; LAvol-A4C 68.2±29.7 cm³, 74.4±32.4 cm³). The agreement between automated and manual measurements performed in high-resolution and handheld TTE was assessed using Bland-Altman analysis (Table 1; Figure 1). 

Conclusions: The application of an automated detector to handheld TTE was feasible and led to measurement values in the same range when compared to human reading in both high-resolution and handheld TTE. The automated detector demonstrated equal performance in HHE and HRE with superior agreement in distances than in volumes. Our results await extension to further echocardiographic parameters and confirmation in larger cohorts but suggest that automated echocardiographic reading might become a valuable tool in patient care.

Funding: This work was funded by a grant from the Bavarian Ministry of Economic Affairs, Regional Development and Energy, Germany (MED-1811-0011).

Table 1    Mean difference ± 95% limits of agreement regarding automated and manual measurements performed in high-resolution standard transthoracic echocardiograms (HRE) and handheld echocardiograms (HHE).

IVS, interventricular septum; LVD, left ventricular diameter; PW, posterior wall;
LA, left atrium; A2C, two-chamber view; A4C, four-chamber-view.
Available in n = 51 participants.