Determining Cardiac Function With and Without an AI Tool

The integration of AI into point-of-care ultrasound (POCUS) is rapidly transforming the landscape of emergency medicine (EM). At the recent American College of Emergency Physicians (ACEP) Scientific Assembly, I had the opportunity to present our team's research on the impact of AI-assisted cardiac function assessment, specifically, the Real-Time EF tool on trainee confidence and workflow in a high-volume emergency department (ED).Figure 1: Real-time EF feature on GE venue.Background: The Evolving Role of POCUS and the Promise of AI POCUS has become an indispensable tool in the emergency physician's diagnostic arsenal, particularly for the rapid evaluation of undifferentiated shock.1-3 One of its most critical applications is the assessment of cardiac function, where timely identification of cardiogenic shock can significantly influence patient outcomes. Traditionally, emergency physicians have relied on qualitative visual estimation of left ventricular ejection fraction (LVEF), a skill that is highly dependent on operator experience and can be challenging for trainees to master.3-5 Recent advances in AI and machine learning have led to the development of tools designed to standardize and streamline cardiac assessment. The Real-Time EF feature, available on the GE Venue ultrasound platform, leverages machine learning algorithms to automatically estimate LVEF from the apical four-chamber (A4CH) view (Figure 1). This tool provides a rapid, color-coded assessment of image quality and delivers an LVEF estimate in seconds, eliminating the need for ECG gating or manual tracing.6,7 Study Objectives: Evaluating AI's Impact on Trainee Performance While previous studies have largely focused on experienced sonographers or have been industry-sponsored, our investigation sought to address a gap in the literature: How does AI-assisted cardiac assessment affect trainees, the medical students and residents who are still developing their POCUS skills?8-10 Our primary objective was to compare trainee confidence in LVEF assessment before and after using the Real-Time EF tool. Secondary objectives included evaluating the tool's perceived usefulness and practicality, as well as the concordance between trainee and AI-generated LVEF estimates. Methods: Real-World Assessment in a Busy Urban ED We conducted a cross-sectional study over a 10-month period in a high-volume, urban emergency department. Twenty trainees, including medical students and residents at various stages of training, participated during their standard ultrasound rotation. Each participant received a brief orientation to basic POCUS principles through a workshop or web-based modules. For each eligible patient presenting with a cardiopulmonary complaint, trainees performed a standard cardiac POCUS exam, obtaining the parasternal long axis, parasternal short axis, apical four-chamber, and subcostal views. Initially, trainees made a qualitative assessment of LVEF (categorized as indeterminate, hyperdynamic, normal, reduced, or severely reduced) and rated their confidence on a five-point Likert scale. They then used the Real-Time EF tool to obtain an AI-generated LVEF estimate and rated their confidence again. The time required to obtain the A4CH view and AI estimate was recorded, and trainees were asked if they found the tool useful in their assessment. Results: AI Significantly Increases Trainee Confidence and Is Widely Regarded as Useful Over the course of the study, 56 POCUS cardiac exams were performed. The majority of trainees were PGY-1 residents (64.3%) and medical students (30.4%), with a smaller percentage of more senior residents. The results demonstrated a statistically significant increase in trainee confidence following the use of the Real-Time EF tool. The mean confidence score prior to using the AI tool was 3.41 (SD 0.83), which increased to 4.14 (SD 0.96) after AI assistance (P < .0001) (Table 1). This improvement was observed across all experience levels, with novice users benefiting most from the additional support. Table 1 - Mean Confidence Levels Pre-and Post-Real-Time EF Tool Utilization Real-time EF tool utilization Mean (SD) confidence level Pre- 3.57 (0.81) Post- 3.94 (1.06) Abbreviation: EF, ejection fraction; SD, standard deviation. In terms of perceived utility, 82.1% of trainees reported that the Real-Time EF tool was helpful in their assessments. The concordance between trainee qualitative assessments and AI-generated LVEF estimates was high, with agreement in 84.3% of cases. While novice users required more time to acquire the necessary cardiac views (mean 27.98 seconds) compared to competent users (mean 11.25 seconds), the tool was accessible and practical for all participants. Discussion: Implications of AI-Assisted POCUS for Trainees and EM Practice Our study demonstrates that AI-assisted tools like Real-Time EF significantly increase trainee confidence in cardiac function assessment, with most users finding the tool helpful and a high concordance between trainee and AI-generated LVEF estimates. While our primary focus was on confidence rather than diagnostic accuracy, these findings suggest that AI can effectively support novice users as they develop POCUS skills in the emergency department. Beyond the immediate educational benefits, the integration of AI into POCUS has the potential to reshape EM practice. For trainees, increased confidence is likely to translate into more frequent and meaningful use of ultrasound, accelerating their learning curve and encouraging earlier, more independent clinical decision-making. As AI tools provide real-time, objective feedback, trainees may feel empowered to incorporate POCUS into routine patient care, potentially leading to earlier recognition of critical conditions and improved outcomes. For practicing physicians, AI offers the promise of greater consistency and efficiency in cardiac assessments, reducing interobserver variability and supporting clinicians across all experience levels. As these tools become more robust and accessible, we may see a democratization of ultrasound expertise, with a broader range of providers able to perform and interpret cardiac POCUS reliably. Looking to the future, the implications extend beyond just increased ultrasound use. As AI continues to lower the barriers to high-quality image acquisition and interpretation, it is conceivable that a wider array of health care professionals could participate in bedside ultrasound, further expanding its reach and utility in the ED. This could lead to more streamlined workflows, enhanced team collaboration, and ultimately, better patient care. In summary, our findings highlight not only the educational value of AI-assisted POCUS for trainees but also its potential to drive broader changes in EM. As technology evolves, AI shows promise in making rapid, reliable cardiac assessment a standard part of emergency care, redefining roles, expanding access, and elevating the quality of care for all patients. Ethics Statement and Disclosures This investigation was approved through the AHN Institutional Review Board and meets all ethical guidelines. Consent was obtained for all study participants. Our authors do not have any disclosures.