Print Friendly, PDF & Email

Why healthcare AI is both an art and a science

This partnership is combining one million images and 22,000 patient studies with a distinct strategy

When the American Heart Association and American Stroke Association updated their ischemic stroke guidelines to expand the treatment window from six to up to 24 hours,[i]  it did more than offer hope to the millions who may be at a heightened risk for this debilitating occurrence. Cardiologists, neurologists and radiologists involved in stroke care say it also means treatment will be based on stroke size and character rather than a strict time cutoff, creating both a challenge and an opportunity.

When a stroke is suspected or has occurred, fast, high-quality CT and MRI scans become even more important.

Enter an AI stroke application being developed[ii] by Partners Healthcare (PHS) and GE Healthcare that the creators say has the potential to help simplify and organize images for stroke detection and treatment in situations where “time = brain.” The data feeding the algorithm being developed comes from a trove of one million images and 22,000 patient studies aggregated and annotated by more than 10 top clinicians and researchers at PHS. Additionally, the teams plan to leverage seven healthcare institutions across the globe to validate the Boston-originated app in development, so that it could be used by clinicians to help treat patients no matter where they live in the world.

“While AI models may be perfectly suited for the environment in which they were trained, sometimes that capability falls off in a new environment,” said Dr. Mark Michalski, ‎executive director of MGH & BWH Center for Clinical Data Science and a leader in the PHS/GE Healthcare collaboration. “[Validation] will be one of the biggest challenges we have as a community, moving forward, to understand how to ensure these models work safely across multiple environments, scanners and tasks.”

Ensuring AI is created and validated with precision is especially important when it comes to stroke. Every 40 seconds, someone in the US has a stroke. The resulting cost of treatment, medicine and missed work days totals $34 billion.[iii] Many localities in the US are unable to provide optimal detection and treatment of stroke due to lack of standardized imaging protocols, gaps in expertise and delays in time and reporting.

These statistics demonstrate that the problem is too great to not get AI right. They also demonstrate why those one million images, 22,000 patient studies and seven validation partners really matter.

“An algorithm will be better at detecting stroke if it sees one million examples versus one hundred,” said Keith Bigelow, Senior Vice President of Edison Portfolio Strategy at GE Healthcare. “But those one million examples must also be grounded in truth – quality data, validated by top clinicians.”

According to Bigelow, successful AI development in healthcare is both a science and an art.

“It’s a science because of the need for large, diverse and quality data sets. That’s a numbers game. But healthcare AI is also an art – we must ensure safe, fair and effective creation and usage, which requires not just strong data but strong strategy, validation and ethics.”

This week at the World Medical Innovation Forum (WMIF) – a global gathering of more than 1,700 senior healthcare leaders established to respond to the intensifying transformation of healthcare – GE Healthcare CEO Kieran Murphy reinforced both the science and the art side of the equation. The company announced that its Edison intelligence platform will integrate with the American College of Radiology AI-LAB, which will allow ACR members and other radiology professionals to more easily develop and seamlessly deploy their algorithms across hospitals and research centers nationally. Additionally, speaking on a panel, Murphy shared the company’s guiding principles for AI development and deployment.

GE Healthcare isn’t alone in publicly committing to ethical AI, but it is one of the first healthcare companies to put forth such guidelines. Thus far, traditional, non-healthcare technology companies have dominated the discussion.

Asked why principles like these are important to highlight now, especially given that healthcare has always been driven by ethics, Dr. James A. Brink, Radiologist-in-Chief at Massachusetts General Hospital, said that the increasingly integral role of AI makes ethics relevant in new ways to the industry.

“With these tools come remarkable new capabilities, leveraging the power of big data as applied to human health and disease,” Dr. Brink said in an interview with Murphy. “However, they are relatively opaque to the end user, in part owing to limited data science knowledge among healthcare workers, and in part owing to the ‘black box’ nature of the algorithms themselves.”

GE Healthcare said it hopes its practical, methodical strategy – focusing not just on data volume but data variety and veracity – coupled with its AI principles, will address the challenges and opportunities that come with AI in healthcare

 

[i] https://www.ahajournals.org/doi/full/10.1161/STR.0000000000000158

[ii]  Technology in development that represents ongoing research and development efforts. These technologies are not products and may never become products. Not for sale. Not cleared or approved by the U.S. FDA or any other global regulator for commercial availability

[iii] https://www.cdc.gov/stroke/facts.htm