Dr. Luciano Prevedello, a radiologist at the Ohio State University Wexner Medical Center (OSUWMC) and chief of the division of medical imaging informatics, appreciates the potential he sees for artificial intelligence in healthcare. He’s well-versed in the evolution of the technology, and over the last three years, he has had the opportunity to more deeply explore the strengths and weaknesses of deep learning applications. Recognizing the importance of the field, under the leadership of Dr. Richard White, chair of Radiology, the department has decided to expand activities in this area and create a lab dedicated to developing and studying AI applications in medical imaging.

In this first part of HealthCare Business News’ conversation with Dr. Prevedello, we’re taking a look at AI’s past, present and future in healthcare, as well as some of the challenges and benefits it faces in the realm of medical imaging.

Although there’s a lot of buzz about developments in AI, and Prevedello shares the excitement, there are some things he sees differently. For one, while it’s a developing story, it’s also an old story. “My view is probably different than some people,” he explained. “A lot of people refer to AI as a futuristic thing, but AI is already happening and has been here for quite some time. AI in the form of machine learning has already been used in many applications in medical imaging.”

For Prevedello, what is different now is that recent developments in machine learning, and more specifically, deep learning, will enable more complex analyses and will broaden the scope of artificial intelligence applications in Medical Imaging. These new techniques can be applied not only to medical images but also to free text clinical reports. While some degree of automation was possible with traditional machine learning tools, they required extensive customization making development very resource intensive. “It was nearly impossible to apply the exact same algorithm to different clinical scenarios”, he explains. Recent techniques are much more generalizable. For example, the same deep learning algorithm can be trained to recognize pneumonia on a chest x-ray or intracranial hemorrhage on a CT of the head. “In the near future, we will start seeing the introduction of new algorithms doing much more sophisticated things. These advancements will not be necessarily readily apparent to us. Sometimes, it just looks as though the applications have become “smarter”. It’s similar to what we’ve seen for the web search industry. In the beginning it was simpler, string-based matching and there were some rule-based algorithms that could connect us to websites. And then Google introduced a different way of dealing with the web, with a lot more sophistication and machine learning behind the scenes,” Prevedello said.

That comparison is important he says, because it seems very similar to what’s happening in healthcare. “While we’ve already had some of these tools for a while, with time, they’re going to become more embedded, more intelligent, to the point where we’ll be asking ourselves a few years from now how we were able to function without them before.”

In time, he believes there will be more diagnosis assistance, but it will require more validation, regulations and tight oversight. A few narrow-focused clinical applications have already been approved by the FDA. Once the level of knowledge and research builds along with the financial interest from technology providers, Prevedello foresees that clinical algorithms will expand much further in their capabilities. “I think soon enough, applications will be used in day-to-day clinical practice to augment our capabilities. We know that computers can extract a lot more information from the pixel data than we can perceive with our naked eyes. This can be used to improve our diagnostic capabilities or even go further to suggest the genetic profile of some tumors to allow more personalized and targeted treatment decisions.”

For all of Prevedello’s optimism for the future of AI, he’s a realist. “Although the evolution and development of AI applications has been happening at a much faster pace than we have imagined in the past, there is still a huge implementation gap. That’s why I think we need to work on standardization of the processes to allow us to easily integrate these tools into the clinical setting. It’s not an easy task to do. It takes a lot of people, a lot of effort. When we see all the news on AI, we have a perception that all these applications will be in our environment pretty soon. But the reality is that it takes a lot of effort to embed them into clinical systems. It will take longer than people are anticipating. “