With further validation, this technology could help 'democratize' knowledge and benefit a wide range of businesses.
LONG BEACH, Calif.—Artificial intelligence (AI) is tasked with detecting important non-coronary data on angiograms that can help modify or enhance treatment plans. It has been suggested that this is possible.
“We tend to miss things we're not looking for,” said Mohammad Alkhouri, MD, PhD (Mayo Clinic School of Medicine, Rochester, Minn.), who presented the study at the Featured Clinical Research Session here at SCAI 2024.
AI has become a hotbed of research in the field of cardiology, with machine-based learning aimed at sniffing out relevant clinical information that might otherwise be overlooked. Recent examples include its use for identification. LVEF is low, Left ventricular systolic dysfunction, Long QT syndromeand coronary artery calcium.
For the AI model, Alkhouli and colleagues handpicked more than 20,000 angiograms from coronary cases performed at the center between 2015 and 2022. The model learned to not just detect occlusions, but to extract data on left ventricular ejection fraction, diastolic dysfunction, and right ventricular dysfunction. , cardiac index from just one or two angiography videos. Alkhouli said these four elements were chosen for the model to be mined because the researchers believed they would be useful to operators in the context of acute cardiac disease.
For predicting ejection fraction ≤ 40% and predicting high filling pressures, the area under the curve (AUC) of the AI model was 0.87 when compared with echocardiography performed within 30 days after angiography. For right ventricular function, the AUC was 0.78 when compared with echocardiography, whereas for cardiac index it was 0.74 when compared with concurrent right heart catheterization.
Although further validation is needed, Alholi said the center is currently incorporating the model into a local cloud in one of its cath labs, “so it can act as a kind of dashboard to spit data back to the operators in the room.” Stated. You can actually use it in a practical way. ”
AI as an assistant
Panelist Yiannis Chatzizisis, MD, PhD, Miller School of Medicine at the University of Miami, Florida, said the results show another way AI could serve as an assistant to busy operators. He said the model essentially represents a “democratization of knowledge” effort, giving everyone at a particular center access to information gleaned from cases performed by their colleagues, and that the He suggested that this is equivalent to extracting every last drop of fruit juice.
Panelist J. Dawn Abbott, MD (Brown University, Providence, Rhode Island) added, “There are so many ad hoc procedures being done right now that this instant feedback is invaluable.” . The real test is whether the use of AI leads to improved patient outcomes, she said.
Alholi said validation will be an important next step for this research, and practical trials are being planned to understand how the use of AI can enhance resource utilization.
He gave the example of a patient with STEMI who has not had a previous echocardiogram and for whom interventions have not been successful. “The model indicates that this patient may be in cardiogenic shock and the filling pressure is elevated. So we need to review that information and treat the patient accordingly. Therefore, either resource utilization increases or there is no need to take additional steps,” he told TCTMD.
