Achieving zero variability echocardiographic analysis in cardio-oncology with AI-powered EchoGo from Ultromics
In November 2021, a paper published in the journal Circulation compared analysis of global longitudinal strain (GLS) in cardio-oncology patients using EchoGo’s artificial intelligence platform with two competitor platforms using traditional semi-automated techniques. It concluded that not only is AI-based strain analysis with EchoGo possible without operator input, but also provides earlier detection and better prognosis with the facilitation of GLS in those patients [1].
Left ventricular (LV) global longitudinal strain (GLS) is a robust LV systolic function measure used to detect subtle chemotherapy cardiotoxicity. However, inter-reader and inter-vendor variabilities compromise the clinical value of longitudinal follow-up of GLS. Artificial intelligence (AI)-based, fully automated measurement of longitudinal strain may be more reliable compared with human interpretation.
The study
The study included 52 transthoracic echocardiographic examinations randomly selected from a Cardio-oncology registry. All subjects received anthracycline-based chemotherapy in 2016-2019. AI-based longitudinal strain was assessed by EchoGo Core using standard 2- and 4-chamber apical views. Two readers verified the myocardium tracing by AI and found no errors. Longitudinal strain results by EchoGo were compared to GLS measured by conventional software (TomTec and Philips QLAB) using standard 3-, 2- and 4-chamber apical views.
The results
The results showed AI-based longitudinal strain analysis was feasible in 51/52 (98%) transthoracic echocardiographic studies. The mean longitudinal strain was -17.3±3.3% for EchoGo, -16.9±2.4% for TomTec and -17.5±3.1% for QLAB. Bland-Altman analysis showed a bias of -0.4 ± 2.7% and 95% limits of -5.7 to 4.9% between EchoGo longitudinal strain and TomTec GLS (Figure 1A). A bias of 0.2 ± 3.3% and 95% limits of -6.2 to 6.6% between EchoGo longitudinal strain and QLAB GLS (Figure 1B) were seen. The bias between TomTec GLS and QLAB GLS was 0.6 ±2.2% (Figure 1C). The inter-reader correlation coefficients of TomTec GLS and QLAB GLS were 0.57 and 0.71, respectively.
This novel AI-based longitudinal strain analysis was feasible in the majority of echocardiograms without any operator input. The bias between EchoGo longitudinal strain and conventional software appears small. AI-based myocardial strain analysis may reduce variabilities and facilitate longitudinal follow-up of GLS in Cardio-oncology patients.
EchoGo from Ultromics is making zero variability a reality, not only for cardio-oncology but heart failure, coronary artery disease, amyloidosis and other disease areas. EchoGo Core and its sister product EchoGo Pro are vendor-neutral and platform-agnostic, with no on-site software required thanks to their cloud-based AI analysis.
Ultromics and EchoGo
Ultromics was founded in 2017, and in 2019, the study that led to EchoGo’s approval by the FDA and gaining CE marking said, for inter-operator variability of continuous variables (EF, GLS, and volumes) when processing the same image clips with EchoGo Core demonstrated no variation was observed [2].
EchoGo uses a one-of-a-kind outcomes-based dataset containing billions of data points from its partnerships with the NHS, the Mayo Clinic in the US and other leading global health systems. Removing manual analysis and human interpretation from the diagnostic pathway saves time and delivers safe and effective results, with zero operator variability. Standardization comes as standard, with the same high precision results all the time, every time. Zero variability is not an aim of EchoGo, it is a fundamental function of the technology.
References:
- Hanif W, Goldberg Y, Taub CC, et al. Abstract 11383: Automated Measurement of Global Longitudinal Strain by Speckle-Tracking Echocardiography in Cardio-Oncology Patients Using Artificial Intelligence. Circulation. 2021;144(Suppl_1).
- 31st ASE Annual Scientific Sessions. American Society of Echocardiography 30th Annual Scientific Sessions. 2020;33:6-A1-A16, B1-B156, 647-778