Artificial intelligence enhanced ECG as a gatekeeper for advanced cardiac imaging in heart failure: a real-world validation study
E Figueiredo, T Branco, L Alves, B Viana, J Goncalves, M Rocha, H Moreira, P Palma, M Vasconcelos, C Sousa, T Pinho, R RodriguesAbstract
Background
Advanced cardiac imaging techniques such as transthoracic echocardiography and cardiac magnetic resonance (CMR) are central to the diagnosis of left ventricular systolic dysfunction (LVSD) but require substantial technical and healthcare resources. Artificial intelligence–enhanced electrocardiography (AI-ECG) has emerged as a widely accessible approach that may optimize referral pathways for advanced imaging. PMcardio® is a commercially available AI-ECG platform capable of automated assessment of left ventricular function, yet its potential role as a gatekeeper for cardiac imaging has not been fully explored.
Objectives
To evaluate the feasibility and clinical impact of using the PMcardio® AI-ECG model as a gatekeeper for advanced cardiac imaging in patients undergoing evaluation for heart failure.
Methods
In this retrospective observational study, 384 patients who underwent ECG, transthoracic echocardiography, and CMR within a 30-day interval were included. Standard 12-lead ECGs were analyzed using the PMcardio® AI-ECG model, which provided a binary classification of left ventricular systolic function (<50% vs ≥50%). CMR served as the reference standard. Using the observed diagnostic performance of the AI-ECG model, we simulated an AI-ECG–guided diagnostic strategy in which referral for advanced cardiac imaging would occur only in patients with a positive AI-ECG result.
Results
LV systolic dysfunction by CMR was present in 112 patients (29.2%). The PMcardio® AI-ECG model demonstrated a sensitivity of 85.7% and a specificity of 85.3% for the detection of LVSD. In the simulated gatekeeper strategy, 136 patients (35.4%) would have been referred for advanced cardiac imaging based on a positive AI-ECG result, corresponding to a potential reduction of 64.6% in imaging utilization. Sixteen patients (4.2% of the total cohort) with LVSD would not have been referred for imaging due to false-negative AI-ECG results. The negative predictive value of the AI-ECG model exceeded 93%, supporting its ability to safely exclude LVSD in a large proportion of patients.
Conclusions
An AI-ECG–guided diagnostic strategy using the PMcardio® platform could substantially reduce the need for advanced cardiac imaging while maintaining a low rate of missed left ventricular systolic dysfunction. These findings support the potential role of PMcardio® AI-ECG as an effective gatekeeper in heart failure diagnostic pathways, improving efficiency and access to care.AI ECG and Imaging pathwayFor image description, please refer to the figure legend and surrounding text.