May 19, 2026
Traditionally, angiograms have been used to evaluate coronary arteries. But what if the same images reveal more than coronary artery disease? That is what Mayo Clinic Cardiovascular Medicine researchers sought to uncover.
The researchers decided to explore deep learning (DL) in detecting right ventricular (RV) dysfunction from routinely acquired cine images during coronary angiography. By using artificial intelligence (AI), the researchers were able to extract further value from the data. The images offered insight into RV function, which they were not designed to do.
Challenging assumptions
Mayo Clinic's innovative research involved training 3D convolutional neural networks (CNN) to identify RV dysfunction from diagnostic angiography of the right coronary artery (RCA). This novel DL algorithm could accurately detect RV dysfunction from routine RCA cine angiography. The AI model's performance improved further after adding electrocardiogram (ECG) data. By combining AI-Cath and AI-ECG models, the researchers built an ensemble AI model. Their findings were published in European Heart Journal — Digital Health.
The final study cohort included 9,849 patients with 10,336 coronary angiograms captured at cath labs within Mayo Clinic campuses in Minnesota, Florida and Arizona. The mean age of patients was 66 years and 36% were women. The angiograms were extracted between 2015 and 2021.
"This is the first study to apply AI to extract RV functional information from angiography and it represents a conceptual shift. It challenges the assumption that we fully understand what our routine imaging is telling us. By training a deep learning model on angiograms that cardiologists interpret every day, we show that these images contain meaningful information about right ventricular function. That opens a new way of thinking about existing data," says Puskar Bhattarai, Ph.D., M.S., a senior data science analyst in Cardiovascular Medicine at Mayo Clinic in Rochester, Minnesota, and a study author. Dr. Bhattarai works in Mayo's Artificial Intelligence (AI) in Cardiovascular Medicine specialty group.
Real-time RV viewing
This becomes more apparent in the clinical environment. "In urgent situations such as acute myocardial infarction, cardiogenic shock or suspected pulmonary embolism, patients arrive in the cath lab quickly, and decisions need to be made even faster. This is when understanding RV function matters most, but we often have little information because there isn't time for a preprocedure echocardiogram," Dr. Bhattarai says. "A tool that screens for RV dysfunction from an angiogram already acquired, without adding time, steps or disruption to workflow, addresses a real and crucial clinical gap that currently has no practical solution."
A key factor of why RV dysfunction is critical in cardiology is because it is often underrecognized. "In emergent scenarios, early or subclinical RV dysfunction can be hemodynamically compensated, so the patient may appear stable while the right ventricle is already under strain," Dr. Bhattarai says. "This creates an important opportunity. The cath lab may represent the first real-time window into ventricular function."
Study highlights
- The model achieved area under the receiver operating characteristics curve (ROC-AUC) of 0.82 and 0.83 for detecting any or significant RV dysfunction, respectively, which is clinically meaningful for a screening tool.
- When the angiogram-based model was combined with an ECG-based model, the AUC for detecting significant RV dysfunction jumped to 0.87. The two modalities carry complementary information about the right ventricle.
- The longitudinal outcomes were over a median follow-up of 6.3 years. Patients classified as false positives, those flagged by the AI model as having RV dysfunction despite normal echocardiograms, had a hazard ratio of 2.33, indicating more than double the mortality risk compared with true negatives.
"Decision curve analysis demonstrated a meaningful net clinical benefit for detecting any RV dysfunction," Dr. Bhattarai says. "This is more than a model that performs well statistically; it has the potential to improve real-world decision-making at the bedside."
Next steps
External validation will be key to demonstrate that this approach performs consistently across different institutions, imaging systems and patient populations. "We want to prospectively evaluate whether identifying RV dysfunction in real time changes clinical decision-making and improves outcomes. That's the pragmatic trial we're currently designing," Dr. Bhattarai says.
The new approach doesn't replace echocardiography. It complements it by providing clinically relevant information when decisions need to be made in real time.
"This study reflects Mayo Clinic's leadership in AI," Dr. Bhattarai says. "Discoveries like this are enabled by the scale of the data, depth of longitudinally linked imaging records across multiple campuses, and a collaborative culture between clinicians and data scientists. Mayo Clinic provides a powerful environment for innovation."
For more information
Rostami B, et al. Deep learning for estimating right ventricular function from routine coronary angiography. European Heart Journal — Digital Health. 2026;7:ztag004.
Refer a patient to Mayo Clinic.