VIDEO: Where will radiology AI be in 5 years?

Dreyer also holds the positions of vice chairman of radiology at Massachusetts General Hospital, chief data science and information officer for the departments of radiology for both Massachusetts General Hospital and Brigham and Women's Hospital, and associate professor of radiology at the Harvard Medical School. 

"There are going to be AI breakthroughs in the technology," Dreyer explained.

He said these breakthoughs in radiology AI in the next few years that will lead to rapid change in the AI market. His top 5 technology advances to watch include:

   1. Development of generalizable AI algorithms that can easily be adapted for multiple uses, rather than programing complete algorithms from scratch. This will help speed AI algorithm development and coming to market.

   2. The development of AI medical imaging virtual development test environments. Dreyer said this will make AI algorithms much faster to build. Current AI models require a lot of infrastructure building that could be replaced by virtual environments for several apps. 

  3. Development of deeper integrations with current PACS and enterprise imaging vendors and sales through only a couple App Store-type market places. He said there is movement away from app stores by some PACS vendors to partnerships to integrate specific apps into their platforms. For wider adoption of AI, it needs to integrate easily across the board and be easy to purchase and use, which he feels requires an App Store model.
   
   4. Autonomous diagnostic AI development and FDA clearance. Dreyer pointed to PAP smear algorithms that work nearly autonomously now, gaining clearance to assess digital microscope slides automatically, without oversight from human pathologists or technologists. 

   5.  Superhuman AI to do things humans cannot. He said this is where AI alone can put together many pieces of data not evident to the human eye to assess the radiomics of images.

"With super human AI, I think that is when you will start to see reimbursement coming in," Dreyer said. He explained this is because the AI technology will offer a means to advance clinical care that is not currently available using current technologies.

   6. AI to screen patients. "I think we will get to the point where he have 'the computer will see you now' notion," Dreyer said. This will be very targeted applications, such as screening patients prior to an MRI.

He said one-off AI diagnostics for acute emergency conditions like pulmonary embolism (PE) is helpful to a care team and speed along the continuum of care. However, human involvement is still required because usually there are a lot of other factors involved in PE that are important to know, or the cause of the patient's condition might have other causes, or there may be incidental findings missed by the AI.  

"If you have a perfect pulmonary embolism detector on CT, that is not the only thing a radiologist looks for on the CT," he said. "Radiologists also look for heart failure and a variety of other things. And usually, it is less often that you find a PE than it is  you find other diagnoses. So just because you have a superhuman PE detector, it is not enough for that algorithm to diagnose that entire exam." 

While is was predicted a few years ago AI would threaten the jobs of radiologists, Dreyer said AI cannot replace real people doing the job anytime soon. He also said there is currently a radiologist shortage and AI will be needed to enhance the capabilities of the radiologist to become more efficient. 

This is part of a 4 video interview series with Dreyer on various aspects of radiology AI. Watch the others:

VIDEO: Development of AI app stores to enable easier access

VIDEO: Overview of radiology AI by Keith Dreyer

VIDEO: Segmenting the Radiology Artificial Intelligence Market by Function