AI Electrocardiogram Model Aids Aortic Dissection Discrimination in Retrospective Study

An artificial intelligence model combining 12-lead electrocardiogram waveforms with patient age and sex demonstrated the ability to discriminate acute aortic dissection beyond demographic factors alone in retrospective chest pain cohorts, according to a study published in BMC Medical Informatics and Decision Making.

Acute aortic dissection represents a life-threatening cause of chest pain that can be difficult to recognise in emergency and clinical settings. Because routine 12-lead electrocardiograms are standardly acquired during initial evaluations for chest pain, researchers investigated whether artificial intelligence analysis of these tracings could provide additional diagnostic information to aid in identifying the condition.

The retrospective two-centre investigation evaluated a total of 27,160 patients from Zhongshan Hospital, Fudan University, who were divided at the patient level into development, tuning, and internal test sets. External evaluation was subsequently conducted using a cohort of 844 patients from the Qingpu Branch. Classification thresholds were established using the internal tuning set and were applied unchanged to both the internal and external evaluation cohorts, while calibration, alert burden, and performance in an aortic computed tomography angiography-verified subset were also assessed.

In the internal test set consisting of 5,433 patients with 30 acute aortic dissection events, the full model achieved an area under the receiver operating characteristic curve of 0.803, with a 95 percent confidence interval between 0.716 and 0.880. In the external cohort of 844 patients with 31 acute aortic dissection events, the full model achieved an area under the curve of 0.857, with a 95 percent confidence interval from 0.778 to 0.925. By comparison, an electrocardiogram-only model recorded areas under the curve of 0.798 internally and 0.842 externally, whereas an age-and-sex model achieved 0.670 internally and 0.553 externally. Furthermore, in a subset of 94 internal test patients verified by aortic computed tomography angiography, the full model maintained an area under the curve of 0.781, with a 95 percent confidence interval between 0.671 and 0.877.

At the tuning-set Youden threshold, the model achieved a sensitivity of 63.3 percent and a specificity of 81.0 percent internally, alongside an alert rate of 19.3 percent. In the external cohort, sensitivity reached 77.4 percent, specificity was 76.9 percent, and the alert rate was 25.1 percent. The researchers concluded that the observed sensitivity and alert burden do not support stand-alone diagnostic use for the tool, emphasising that prospective evaluation remains necessary to determine whether model-guided clinical reassessment can improve diagnostic decisions and patient safety.

Subject of Research: Artificial intelligence electrocardiogram analysis for acute aortic dissection

Article Title: Development and external validation of an AI-enabled 12-lead electrocardiogram model integrating age and sex for acute aortic dissection

Article References: Zhang, H., Sun, Z., Chen, T., Liu, M., Luo, R., Wang, M., Liu, H., & Jiang, H. (2026). Development and external validation of an AI-enabled 12-lead electrocardiogram model integrating age and sex for acute aortic dissection. BMC Medical Informatics and Decision Making. https://doi.org/10.1186/s12911-026-03951-x

Image Credits: AI Generated

DOI: 10.1186/s12911-026-03951-x

Keywords: acute aortic dissection, artificial intelligence, 12-lead electrocardiogram, chest pain, diagnostic modeling, external validation, clinical informatics, retrospective study, Development, external, validation, AI-enabled