Pre-Hospital Artificial Intelligence-Guided, Focused Echocardiography in Patients with Acute Chest Pain for Diagnosis of Acute Coronary Syndrome

Abstract

Background: Acute chest pain is a common emergency with only 10–20% of cases attributable to acute coronary syndrome (ACS). Rapid and accurate pre-hospital diagnosis remains challenging, particularly for non-ST elevation ACS, where ECG findings may be inconclusive. AI-guided focused cardiac ultrasound (FoCUS) using handheld devices offers a potential solution by enabling immediate functional cardiac assessment. The aim was to investigate the feasibility and diagnostic performance of pre-hospital AI-guided FoCUS for detecting ACS in patients with acute chest pain. Methods: In this single-center, prospective pilot study, 75 patients with acute chest pain were enrolled. FoCUS examinations were performed by experienced sonographers (72%) and EMS paramedics (28%) using AI-guidance for obtaining the apical 4-chamber (AP4CH), apical 2-chamber (AP2CH), and apical 3-chamber (AP3CH) views. The quality of the obtained images was assessed, and quantitative measurements—including left ventricular ejection fraction (LVEF) and global longitudinal strain (GLS)—were analyzed. Diagnostic performance was subsequently evaluated using ROC curve analysis. Results: At least one apical view was acquired in 91% of patients, with sonographer achieving higher acquisition rates than paramedics (96% vs. 67% for the AP4CH view). Complete acquisition of all apical views was achieved in 67% of cases (83% vs. 24%), and image quality was high across views, with median scores ranging from 83% to 100%. GLS yielded an AUC of 0.76 (89% sensitivity, 56% specificity) and LVEF yielded an AUC of 0.65 (75% sensitivity, 73% specificity). In patients with intermediate to high HEAR-scores (>3), lower LS-AP4CH values were associated with ACS. Conclusion: Pre-hospital AI-guided FoCUS is feasible and shows promise for ACS detection, although quantitative parameters do not yet outperform established clinical scores. Enhanced training and further refinement of AI algorithms are needed before widespread implementation.