AI spots critical heart defects missed by the human eye
Researchers from the United States and Japan have developed an advanced artificial intelligence (AI) algorithm for identifying signs of atrial septal defect (ASD), an adult congenital heart disease that often requires surgery, in electrocardiograms (ECGs). The group shared its research in eClinicalMedicine, highlighting the AI model’s long-term potential to make ASD screening a reality.[1]
“If we can deploy our model on a population-level ECG screening, we would be able to pick up many more of these patients before they have irreversible damage,” corresponding author Shinichi Goto, MD, PhD, a cardiologist with the division of cardiovascular medicine at Brigham and Women’s Hospital, said in a prepared statement.
Goto et al. used data from more than 80,000 patients originally seen at one of three hospitals to build their deep learning model. Each patient had originally undergone an ECG and an echocardiogram. Researchers then put their algorithm to the test, using it to evaluate the ECG results of a general population that was not specifically being screened for ASD. Overall, the new-look AI model correctly identified ASD 93.7% of the time—significantly higher than human specialists, who identified ASD 80.6% of the time when looking at the same test results.
“It picked up much more than what an expert does using known abnormalities to identify cases of ASD,” Goto said. “The model's performance was retained even in the community hospital's general population, which suggests that the model generalizes well.”
Another key takeaway from the group’s research was the consistent performance of its AI model. The algorithm was found to be effective across a number of subgroup analyses focused on age, sex, body mass index, the presence/absence of atrial fibrillation and the presence/absence of ECG abnormalities.
Considering its widespread utilization and relatively low costs, ECGs represent an effective way to screen for ASD and other critical heart defects. Combine those factors with the success of the research team’s new algorithm and it’s easy to see why the group is considering the potential of population-level screening.
“Perhaps this screening could be integrated into an annual primary care physician appointment or used to screen ECGs taken for other reasons,” Goto said in the same statement.
Read the full analysis here.
Michael has more than 18 years of experience as a professional writer and editor. He has written at length about cardiology, radiology, artificial intelligence and other key healthcare topics.