UC Berkeley AI Reveals Hidden Heart Patterns to Predict Cardiac Death Risks
Elektrocardiogram scans analyzed by artificial intelligence reveal hidden electrical patterns that predict fatal cardiac events better than standard hospital measurements.
- Deep learning models trained on hundreds of thousands of EKG records identify individuals at high risk for sudden cardiac death who are routinely missed by conventional screening methods.
- The artificial intelligence algorithm detected a patient group facing a 7 percent annual death rate, compared to a 4.6 percent rate identified by traditional left ventricular ejection fraction measurements.
- Researchers published these findings in Nature, opening new pathways for preventive cardiology and implantable defibrillator allocation.
More than 300,000 people in the United States die from sudden cardiac arrest each year, according to data cited by ScienceDaily. The emergency occurs when the heart’s electrical system fails, stopping blood flow.
While implantable defibrillators can prevent these deaths, doctors struggle to identify which patients need them most. Researchers at UC Berkeley identified a previously overlooked warning signal in routine heart scans using an artificial intelligence system trained on more than 440,000 electrocardiograms collected in Sweden, ScienceDaily reported.
“One thing that makes the problem very tragic, but also very well suited for AI, is that we have the cure for this problem,” said Ziad Obermeyer, the study’s lead author and an associate professor at UC Berkeley’s School of Public Health.
According to ScienceAlert, sudden cardiac arrest is believed to cause up to 15 percent of all deaths globally and is estimated to take up to 300,000 American lives annually. The research team evaluated the technology using thousands of additional patient records from the United States and Taiwan to test whether the electrical patterns could reliably identify people at heightened risk across different populations.
Comparing AI screening with conventional ejection fraction
Current screening methods largely focus on left ventricular ejection fraction, an ultrasound measurement of how much blood the heart pumps with each contraction. Using traditional measurements, doctors identify a patient group with an annual sudden cardiac death rate of 4.6 percent, ScienceDaily reported.
By contrast, ScienceAlert said that left ventricular ejection fraction flagged 1.9 percent of the Swedish sample as high-risk, while the artificial intelligence system identified a higher-risk group comprising 2.2 percent of the sample with an annual death rate of 7 percent.
| Screening Method | Percentage of Sample Flagged | Annual Sudden Cardiac Death Rate |
|---|---|---|
| Traditional Left Ventricular Ejection Fraction | 1.9 percent | 4.6 percent |
| AI Electrocardiogram Classifier | 2.2 percent | 7.0 percent |
The system pinpointed a larger group of individuals facing increased danger, which included thousands of patients who would typically not be flagged by current clinical protocols. Left ventricular ejection fraction failed to identify 86.1 percent of the high-risk patients detected by the artificial intelligence, according to ScienceAlert. Among high-risk patients identified by the algorithm, those with implanted defibrillators had a 54.4 percent lower sudden cardiac death rate than statistically expected, ScienceAlert reported.
Researchers assembled medical datasets over ten years
Developing the algorithm required three independent collections of medical data gathered from Sweden’s healthcare system, a San Diego hospital system, and Taipei. Assembling the datasets took approximately 10 years and involved contributions from Dandelion Health and Nightingale Open Science, ScienceDaily reported.
“Good AI starts with good data,” Ziad Obermeyer, the study’s lead author and an associate professor at UC Berkeley’s School of Public Health, remarked to ScienceDaily.
When the researchers used a second model to visualize the associated electrical patterns, they uncovered a previously undescribed waveform feature, ScienceAlert reported. To integrate the algorithm into hospital EKG databases, Obermeyer is currently working with medical institutions across the United States, Taiwan, and Sweden.
Physicians would contact patients whose records receive a high-risk assessment and offer them a patch that continuously records electrical activity to provide clues about the underlying biological processes.
“Medical decisions are really hard, and I think that’s why AI is so exciting for me,” Obermeyer told ScienceDaily.
Ziad Obermeyer’s path to academic research began in Beirut, Lebanon, before he moved to Cambridge, Massachusetts. Prior to his career in medicine, he worked as a consultant for McKinsey & Company, where he advised global health and pharmaceutical clients across Tokyo, Geneva, and New Jersey. After earning his Doctor of Medicine in 2008, he practiced as an emergency physician at the Fort Defiance Indian Hospital on the Navajo Nation in Arizona.
Disclaimer: The information provided in this article is for educational and scientific communication purposes only and does not constitute medical advice. Always consult with a qualified healthcare provider regarding any medical condition, diagnosis, or treatment plan.