Before his team reopens a blocked artery in a heart-attack patient, Dr. Ankur Kalra wants to know one vital thing: will the heart muscle itself start bleeding? Now, thanks to an artificial intelligence tool his team built at Upstate Medical University in Syracuse, New York, he can finally see the answer coming before it happens.
Kalra, a cardiologist who leads the cardiac catheterization labs and runs the cardiology department at Upstate, co-led a research team that created a scoring system using explainable artificial intelligence, or XAI. "Explainable" is the key word. Most AI acts like a black box — it gives you an answer but won't tell you why. This tool is different: it shows doctors exactly which patient details are driving each prediction. "This is the kind of AI we need for accurate, interpretable and usable data at the point of care," Kalra said.
The problem it tackles is a dangerous one. A heart attack happens when a blocked artery stops blood from reaching the heart muscle. Doctors treat it by restoring blood flow, a procedure called reperfusion. But in about 40% of patients with the most severe kind of heart attack — known as STEMI — that restored blood flow causes the damaged muscle to bleed internally. This is called intramyocardial hemorrhage, or IMH, and it is the most serious form of heart muscle injury. It raises the risk of heart failure and death.
The new scoring system is built for the real world of the cath lab, where cardiologists work while a patient is on the table. Before reopening the blocked artery, the score could flag who is at high risk. In the future it might help doctors decide who needs closer monitoring, who should get a cardiac MRI, and even who could join a clinical trial of a drug called dexrazoxane that is being tested to reduce the damage.
For high-risk patients, those findings could lead to incremental therapy with that intravenous drug. Dexrazoxane is still investigational, but Kalra is presenting phase 2a results in Munich on August 31 at the European Society of Cardiology Congress, with the full study expected in one of the top cardiology journals.
The research, published in JACC: Advances, was a large team effort, with partners at Indiana University School of Medicine, the University of Toledo, Northern Ontario School of Medicine University, and the Cleveland Clinic. Kalra says the findings could eventually become an electronic health record calculator that flags at-risk patients automatically.
With more than 800,000 heart attacks every year in the United States alone, a warning that helps doctors intervene before the damage deepens is more than a clever algorithm. It is a way to give patients a better chance at surviving — and recovering — from the very treatment meant to save them.