Artificial intelligence continues to evolve in cardiology
Click Here to Manage Email Alerts
Artificial intelligence continues to affect cardiology with improved capabilities to diagnose certain conditions such as atrial fibrillation, and research is underway to learn more about its use in disease management, a presenter said.
Although ECG watches were patented in the early 1990s, smartwatches of today are different because of lower manufacturing costs, changes in the regulatory landscape, AI and smartphone-based data transfers, Mintu P. Turakhia, MD, MAS, associate professor and executive director of the Center for Digital Health at Stanford University School of Medicine and a Cardiology Today Next Gen Innovator, said in his presentation at the Scientific Session and Exhibition of the American Society of Nuclear Cardiology.
The use of wearables today has increased due to more people having smartphones, with 81% of the population worldwide owning a smartphone. Approximately 20% of people in the U.S. now own a consumer wearable, which is increasing annually, according to the presentation.
People often track several metrics using consumer wearables, including heart rate, BP and blood glucose. For AF, there has been a shift from measuring heart rate to assessing irregular rhythms, which has led to wearables notifying patients when these occur. The FDA has regulated and even approved some of these capabilities, especially as wearables have entered the medical space.
“There’s a whole theme here in a regulatory cascade from consumer-focused wellness device all the way to a medical-grade diagnostic at home,” Turakhia said during the presentation.
The Apple Heart Study was a validation study to see whether AF can be identified using a smartwatch. Researchers found that 0.5% of participants received an irregular pulse notification, “which doesn’t mean you have AF, but it does tell you that we were not overwhelming the health care system with this low frequency rate across a follow-up period up to about a year and a half,” Turakhia said during the presentation. The study also found that 34% of participants who received an ECG patch after the notification from the Apple Watch had AF. The watch had a positive predictive value of 0.84, which “exceeds the performance of many insertable cardiac monitors that are used to diagnose AF, which have long been considered the gold standard,” Turakhia said.
A similar study was the HUAWEI Heart Study, which was presented and published in 2019. The notification rate was 0.2% in approximately 188,000 participants. Turakhia said the results may have been skewed due to the age of participants in the study.
There are also devices that are not linked to a watch such as the KardiaMobile 6L (AliveCor) device, which can measure six leads, according to the presentation.
Many advancements related to wearables and other consumer products have been made due to AI. This has led to the development of algorithms that can often outperform humans, according to the presentation. These algorithms can be used for workflow enhancement and to diagnose patients based on imaging, for example.
AI may have potential in nuclear imaging such as polar map assessments of cardiac perfusion for coronary distributions. Unfortunately, this had low sensitivity and specificity, according to the presentation.
“Perhaps the discussion might be: Does this represent the ceiling or is there more to be done,” Turakhia said during the presentation. “They actually had a reasonable amount of training data here. This may, in many ways, shows the limitations that you have of a lot of these algorithms where you can only do so well.”
More interest is being placed on reframing AF and stroke risk using patterns of AF, according to the presentation. This may lead to better risk stratification than traditional methods by taking dense data in untreated patients to assess the risk for stroke and other events.
Researchers are also looking at existing diagnostics and looking for new things such as left ventricular dysfunction in ECGs and near-term risk for AF with sinus rhythm ECGs, according to the presentation. New applications are also being assessed in HF with the use of Bluetooth low-energy implantable devices that transmit data through smartphones.
Despite the benefits of using this technology, there are some challenges, according to the presentation.
“A lot of deep learning is based on assumptions inherent to the data acquisition that can be hijacked,” Turakhia said during the presentation.
Other issues include false positives and the lack of devices to aid in disease management. Turakhia said he and colleagues are currently designing a trial to assess the use of a smartwatch for as-needed anticoagulation during prolonged periods of AF. The Heartline Study is assessing the use of an Apple Watch in older patients to detect AF.
In the future, the focus may shift from wearables to noncontact technology, according to the presentation. For example, photos of a patient’s face to assess blood flow may help detect AF, but the issue does not lie with its effectiveness, but rather its deployment due to privacy issues.
“Ultimately, the role for AI is to scale the ability to disintermediate the site of the image acquisition with the need for a specialist and may need to afford some independence as well,” Turakhia said during the presentation.
References:
- Guo Y, et al. J Am Coll Cardiol. 2019;doi:10.1016/j.jacc.2019.08.019.
- Perez MV, et al. N Engl J Med. 2019;doi:10.1056/NEJMoa1901183.
Collapse
Turakhia MP. Cardiac applications of artificial intelligence approaches in wearable devices. Presented at: Scientific Session and Exhibition of the American Society of Nuclear Cardiology; Sept. 25-26, 2020 (virtual meeting).
Click Here to Manage Email Alerts