Machine learning for the real-time assessment of left ventricular ejection fraction in critically ill patients: a bedside evaluation by novices and experts in echocardiography

Abstract

Background Machine learning algorithms have recently been developed to enable the automatic and real-time echocardiographic assessment of left ventricular ejection fraction (LVEF) and have not been evaluated in critically ill patients. Methods Real-time LVEF was prospectively measured in 95 ICU patients with a machine learning algorithm installed on a cart-based ultrasound system. Real-time measurements taken by novices (LVEFNov) and by experts (LVEFExp) were compared with LVEF reference measurements (LVEFRef) taken manually by echo experts. Results LVEFRef ranged from 26 to 80% (mean 54 ± 12%), and the reproducibility of measurements was 9 ± 6%. Thirty patients (32%) had a LVEFRef < 50% (left ventricular systolic dysfunction). Real-time LVEFExp and LVEFNov measurements ranged from 31 to 68% (mean 54 ± 10%) and from 28 to 70% (mean 54 ± 9%), respectively. The reproducibility of measurements was comparable for LVEFExp (5 ± 4%) and for LVEFNov (6 ± 5%) and significantly better than for reference measurements (p < 0.001). We observed a strong relationship between LVEFRef and both real-time LVEFExp (r = 0.86, p < 0.001) and LVEFNov (r = 0.81, p < 0.001). The average difference (bias) between real time and reference measurements was 0 ± 6% for LVEFExp and 0 ± 7% for LVEFNov. The sensitivity to detect systolic dysfunction was 70% for real-time LVEFExp and 73% for LVEFNov. The specificity to detect systolic dysfunction was 98% both for LVEFExp and LVEFNov. Conclusion Machine learning-enabled real-time measurements of LVEF were strongly correlated with manual measurements obtained by experts. The accuracy of real-time LVEF measurements was excellent, and the precision was fair. The reproducibility of LVEF measurements was better with the machine learning system. The specificity to detect left ventricular dysfunction was excellent both for experts and for novices, whereas the sensitivity could be improved. Trial registration: NCT05336448. Retrospectively registered on April 19, 2022.

Artificial Intelligence (AI) versus POCUS Expert: A Validation Study of Three Automatic AI-Based, Real-Time, Hemodynamic Echocardiographic Assessment Tools

“…Another study compared the auto IVC tool in the GE Venue™ device to measurements by POCUS experts and POCUS novices with good results; however, only a five-patient study population was examined [ 26 ]. For EF, a study which studied the agreement between POCUS experts, POCUS novices and the GE auto EF tool was published with good results [ 27 ]. For VTI, one study studied the agreement between the LVOT-VTI automated tool and manual measurements made by the Venue™ device.…”

Diastology in the intensive care unit: Challenges for the assessment and future directions

“…64 Much literature already exists regarding systolic function and fluid status assessment using machine learning algorithms applying automatic measurements of LVEF and left ventricular outflow tract velocity time integral. [65][66][67] Considering all the pitfalls described for the assessment of DD in critically ill patients, we consider a more holistic approach to the dias-tolic function in the ICU more valuable. Experienced clinicians should integrate the hemodynamic information (based not only on echocardiographic variables) with other data from like lung ultrasound (LUS), the venous excess ultrasound (VExUS), and the ventriculo-arterial coupling, 68,69 to better stratify critically ill patients according to their pathophysiologic profile and prognostic impact.…”

“…Hence, we think that it is likely that the current cardiology guidelines will be at some point revised with the chance to introduce PALS or other strain‐based parameters describing LA function, and possibly with a contribution from artificial intelligence 64 . Much literature already exists regarding systolic function and fluid status assessment using machine learning algorithms applying automatic measurements of LVEF and left ventricular outflow tract velocity time integral 65–67 . Considering all the pitfalls described for the assessment of DD in critically ill patients, we consider a more holistic approach to the diastolic function in the ICU more valuable.…”

Use of Artificial Intelligence to Assess Cardiac Function by Echocardiography: Systematic Review of the State of the Art

“…Much of the literature focuses on AI interpretation or analysis of already acquired images, but there is growing evidence regarding the use of AI technology for image acquisition by non-experts. 5,[19][20][21][22][23][24] Automated technology can guide a non-expert examiner by recognizing incorrect images and the need for additional views or enhancements. It can also help standardize image acquisition and measurements.…”

“…1,6 In this way, it would be a tool used to optimize training and echocardiographic studies, 5,25,26 with the possibility of acquiring images of diagnostic quality comparable to those acquired manually by experts. 5,19,20,[22][23][24] The technology is applicable in clinical practice, in emergencies, and in remote areas with scarce resources and few experienced professionals. 5,21,25,27 Integration with clinical data…”