Background <p>Artificial intelligence (AI)–enabled electrocardiograms (ECGs) extracted from hospital electronic medical record (EMR) systems can accurately predict reduced left ventricular ejection fraction (LVEF). However, this limits the applicability to hospital settings. The widespread availability of smartphones presents an opportunity to extend AI-ECG–based cardiac screening to non-hospital settings. The aim of this study was to compare the diagnostic performance of mobile phone ECG images with the standard EMR-based ECG image method.</p> Methods <p>In this prospective validation study, ECG images from 86 patients were analysed using EMR-derived ECG images and photographs of printed ECGs captured using mobile phones. Both inputs were analysed using the same previously validated deep learning model. Echocardiography-derived EF (Ejection Fraction) served as the reference standard. Model performance was assessed using sensitivity, specificity, predictive values, accuracy, receiver operating characteristic area under the curve (ROC-AUC), and precision–recall AUC (PR-AUC), with thresholds selected using Youden’s index.</p> Results <p>Both methods demonstrated identical sensitivity (0.89) and high negative predictive value (NPV) (0.96). Both the EMR-derived and mobile phone–acquired ECG models identified the same number of true positive cases (<i>n</i> = 17) and the same number of false negatives (<i>n</i> = 2). Compared with the EMR images, the mobile photographs showed a slight reduction in specificity (0.75 vs 0.82) and accuracy (0.78 vs 0.84). The ROC-AUC (0.86 vs 0.89) was also similar.</p> Conclusions <p>AI-based LVEF prediction from mobile phone–captured ECG photographs demonstrates performance comparable to EMR-based ECG images, with preserved sensitivity and excellent NPV. This approach enables predicting EF from AI-ECG using mobile phone images and has the potential for screening heart failure in resource-limited and community-based settings.</p>