Newswise — INDIANAPOLIS — Researchers at Indiana University School of Medicine have developed and tested a six-point scoring system that can help identify patients at high risk of internal bleeding in damaged heart muscle after a severe heart attack using explainable artificial intelligence (XAI). Unlike standard AI models that don’t show how conclusions are reached, XAI allows clinicians to identify and discern the clinical factors behind the prediction.
Results of the study were published in JACC: Advances and led by scientists from the Medical Imaging Research Institute Cardiovascular Imaging Research Center at IU School of Medicine.
Intramyocardial hemorrhage or IMH is a life-threatening complication of a heart attack that can occur after doctors restore blood flow through a blocked artery. IMH, the most serious form of heart muscle injury, affects about 40% of patients treated for ST-segment elevation myocardial infarction, or STEMI, a severe type of heart attack, and increases the risk of heart failure and death.
The scoring system is designed for interventional cardiologists to use in cardiac catheterization, or CATH labs before reopening a patient’s blocked artery.
Scoring system can predict IMH
Researchers found that the scoring system could accurately predict IMH using clinical information already available during cardiac catheterization at the patient’s bedside. They were also the first researchers to successfully apply medical use of a traceable AI tool — called Superposable Neural Networks, or SNN. The score uses three measurements obtained in the catheterization laboratory from an electrocardiogram and angiography. Researchers converted the model into a six-point score. A score of 4 or higher are classified as having a high risk of IMH. Those with a score of 3 or lower are classified as having a low risk.
“The significance of this work is that explainable AI does not simply make a prediction; it shows the reasoning behind the prediction, without needing to wait for a cardiac MRI to diagnose IMH,” said Khalid Youssef, PhD, research professor of radiology and imaging sciences at IU School of Medicine and the lead author of the study. “It is a practical, interpretable score that can be calculated before revascularization using data already available in the CATH lab.”
Currently, a specialized cardiac MRI scan known as T2* is the standard method for detecting IMH. However, the scan is often performed 48 to 72 hours after the blocked coronary artery has been opened. By that time, a heart muscle injury could have already occurred and place a patient at a higher risk of IMH or death.
The new scoring system offers a proactive approach by estimating a patient’s risk before blood flow is restored. It is not intended to delay treatment or replace a physician’s judgment. Instead, the scoring provides interventional cardiologists a way to assess risk and adjust care swiftly.
“The central value of this work is that explainable AI can be accurate, interpretable and clinically usable,” said co-author Keyur Vora, MD, director of clinical trials at the MIRI Cardiovascular Imaging Research Center at IU School of Medicine. “Beyond acute emergency, identifying patients at high risk of severe heart muscle injury gives clinical cardiologists the clarity to proactively tailor therapies and improve long-term outcomes.”
Cardiac MRI identified IMH in 142 of the study’s 288 heart attack patients. The model was more than 84% accurate in identifying patients at risk before their blood flow was restored. Researchers said the results demonstrate that the approach is feasible and shows strong initial performance. Larger future studies in development will be used to confirm the findings and determine how the score may guide clinical decisions to improve patient outcomes.
Potential impact for cardiology and medical AI
Researchers said the score could eventually help clinicians assess risk in real time during emergency angiography, when providers check for blockage within the heart arteries, or it could help identify patients who need closer monitoring after a blocked artery is reopened. The score may also help doctors decide which patients need a cardiac MRI and which may qualify to participate in a clinical trial aimed in reducing damage from IMH.
The XAI approach could potentially be adapted for other areas where both accuracy and trust are essential, including critical care, oncology, neurology, medical imaging and clinical trial design.
“This is the kind of AI we need for accurate, interpretable and usable data at the point of care,” said structural heart interventional cardiologist Ankur Kalra, MD, director of cardiac catheterization laboratories and chief, Division of Cardiology, Department of Medicine at State University of New York, Upstate Medical University. “The SNN method allows interventionalists to see exactly which factors are driving the prediction, rather than being asked to trust a black box.”
A multidisciplinary research effort
Five universities participated in the explainable AI scoring clinical study, with a range of expertise.
“Our study required a deeply collaborative, multidisciplinary team because it encompassed cardiovascular medicine, advanced imaging and artificial intelligence — and practical emergency intervention expertise,” said senior author Rohan Dharmakumar, PhD, executive director of the Medical Imaging Research Institute and vice chair of research at the Department of Radiology and Imaging Sciences at the IU School of Medicine. “Interventional cardiology collaborators brought the clinical and procedural expertise needed to identify which variables are available before reperfusion and which variables would be meaningful in a real CATH lab workflow.”
Collaborators included the Cardiovascular Imaging Research Center and the Krannert Cardiovascular Research Center at IU School of Medicine; the University of Toledo College of Medicine and Life Sciences; Northern Ontario School of Medicine University; Cleveland Clinic; and State University of New York Upstate Medical University.
