Healthcare providers continue to adopt AI at a rapid rate, with a majority reporting that they have increased their tech spending over the past year.
The new generation of healthcare AI innovation is promising, but there are some problems that remain when it comes to AI, such as data generalization and the high cost of adopting new tools, pointed out Dr. Arash Harzand during an interview this month at the Heart Rhythm Society’s recent HRX conference in Atlanta.
Dr. Harzand is a professor of cardiology at Emory University in Atlanta, as well as chief health officer at the VA’s Office of Healthcare Innovation & Learning.
When evaluating the new class of AI tools emerging for the cardiology space, he believes that those involving computer vision will have the most impact.
“Computer vision means how we use AI to help guide and interpret diagnostic studies and really any kind of imaging, whether it’s CT, MRI, echocardiography or ultrasound,” Dr. Harzand explained.
About half of FDA approvals in the AI space were granted to tools designed for medical imaging, he added. Some companies that sell computer vision-based tools include Aidoc, PathAI and Qure.ai.
These tools can analyze images much faster than humans, and they often promise to improve diagnostic accuracy by detecting subtle patterns or anomalies in medical images. While AI may have the potential to bring these benefits, the cost of adopting new tools is often a major hurdle, Dr. Harzand noted.
“These things always have a funny habit of actually making costs higher — because it’s healthcare, and anytime you have something new, there is always an upfront cost,” he stated.
Data generalization is another major challenge when it comes to AI, Dr. Harzand pointed out. Healthcare AI tools are typically trained on large datasets, and it’s important to make sure that this training data matches the demographics of the patient population on which the technology is being used.
Take the VA for example — its patient population is very different from the general public. More than 90% of VA patients are men, and many of them were exposed to toxic chemicals like Agent Orange, which can lead to a higher likelihood of having diseases like diabetes, Dr. Harzand said.
Additionally, racial and ethnic demographics look different at every VA medical center — and at every hospital, for that matter — he added. Accounting for these differences and making sure that AI is well-suited to help treat the unique population at hand is a key step in determining the success of healthcare AI tools, Dr. Harzand noted.
Photo: metamorworks, Getty Images
