AI-based system can flag false positives for AF from implantable loop recorders
Click Here to Manage Email Alerts
An artificial intelligence-based ECG analysis solution reduced the rate of false-positive atrial fibrillation detection in patients using implantable loop recorders, researchers reported.
The system (ECG analysis solution, Cardiologs) eliminated as many as two-thirds of false positives from implantable loop recorders and boosted positive predictive value as high as 75%, the researchers wrote in JACC: Clinical Electrophysiology.
Assessing false positives
“To some extent, diagnosing atrial fibrillation is based on pattern recognition, but to some extent, it’s based on being able to distinguish it from other causes of an irregular heart rhythm such as the presence of frequent atrial from ventricular ectopy,” Suneet Mittal, MD, director of electrophysiology and medical director of the Snyder Center for Comprehensive Atrial Fibrillation at Valley Hospital in Ridgewood, New Jersey, told Healio. “I can do that pretty easily when I have access to a 12-lead ECG. However, loop recorders have some challenges. They only give you one lead of data. The segments they acquire are short. It’s often hard for them to distinguish atrial fibrillation from sinus rhythm with frequent atrial from ventricular ectopy. So they can often falsely detect atrial fibrillation. They are tuned in a way so that they don’t miss an episode of atrial fibrillation. Any time you fine-tune something to be very sensitive, then you compromise specificity and can introduce false positives, especially when the episode in question is of a shorter duration.”
The researchers analyzed 425 patients (mean age, 69 years; 62% men) with cryptogenic stroke, syncope or known AF who were using an implantable loop recorder (Reveal LINQ, Medtronic). Using implantable loop recorder readings, the researchers documented which of 1,500 AF episodes were true or false and calculated a positive predictive value. They then applied the AI-based solution and assessed its effect on the positive predictive value.
“If you train a neural network with enough images, it should be able to see what a physician is seeing,” Mittal told Healio.
The researchers excluded 1.1% of ECG readings because they were uninterpretable. Of the remainder, 53.9% were determined to be an actual arrhythmia.
The positive predictive value rose from 53.9% (95% CI, 51.4-56.5) to 74.5% (95% CI, 71.8-77) after the researchers used the AI-based solution (P < .001). The increase in accuracy was driven by better interpretation of AF episodes of 30 minutes or less, Mittal and colleagues wrote.
‘This represents the future’
“Currently available AI-based technologies can successfully whittle away these false positives,” Mittal said in an interview. “In our study, two-thirds of these false-positive episodes could be whittled away. To some extent, this represents the future. You make sure [a technology] does not miss any episodes of atrial fibrillation, and then you let a filter run and give the physicians only the most accurate and actionable information. That’s a win-win for everybody.”
The most common reason for a false positive was premature atrial or ventricular contractions or overdetection of noise. There were very few cases where AI considered a true episode to be false, according to the researchers.
“These ECG data can also come from external devices like watches and smartphone apps,” Mittal told Healio. “In the future, this AI technology can probably be used on all of these types of signals to give providers the truest information possible for the care of their patients.”
For more information:
Suneet Mittal, MD, can be reached at mittsu@valleyhealth.com.
Collapse
Click Here to Manage Email Alerts