Medical AI is On the Scene, Trying to Make Gains
The potential of AI to have wide impact on personal healthcare is high, but challenges around adequate volume and quality of data handicaps projects and studies
By John P. Desmond, Editor, AI in Busines
The use of AI in healthcare has some wins, helping in the rapid development of COVID-19 vaccines most notably. However, medical AI is proving challenging to apply to many areas, including personal healthcare. One observer sees the potential of medical AI to do much more to take care of us.
“AI may have the potential to see what humans cannot and provide a level of care that is otherwise beyond our reach,” stated Peter Wayner, technology writer, book author, in a recent account in VentureBeat.
Medical professionals see the potential of medical AI to be useful to them as high, and the associated risks are seen as high as well, according to survey results. The trick is to exploit the abilities of AI while limiting its potential harm.
The author cites the following challenges:
Imperfect sensors;
Chaotic systems;
Privacy;
Limited knowledge;
Cautious approach;
Tight regulation
“In many cases, sick people have more complex and dysfunctional systems that are harder for algorithms to analyze because they are not behaving normally,” Wayner stated.
Moreover, knowledge about human bodies in many areas is still a mystery, “Sometimes we don’t even know the right questions to ask,” he stated.
On the opportunity side, medical AI has the potential to, for example, develop better sensors that “can identify small changes with better precision than normal human perception,” and that will stay on as long as electric power is available. A medical AI system need not rely on the presence of a human caregiver, but can act as an assistant for the healthcare worker.
“The technology can offer advice to humans, who decide how much of the advice to accept,” Wayner stated.
This approach enables AI to help in a number of areas, including research, training, professional support, patient engagement, remote medicine and hospital care.
Major companies making investments to grow in the medical AI area include: Oracle Microsoft, Amazon, Google and IBM, offering a cautionary tale. IBM recently sold its IBM Watson Health assets to Francisco Partners, which has launched Merative.
Many, many startups are exploring the use of AI to transform healthcare. Wayner cites a few: Medical Harbor and RadLogics, offering platforms to help radiologists capture and interpret images; Molecular Devices and PathAI, using AI to help pathologists, such as by automating repetitive tasks; and Roam Analytics and Sopris, focusing on capturing and storing medical records with the help of AI, in the hopes of increasing accuracy.
Access to Needed Healthcare Data is a Challenge
A recent study conducted at the UK’s Alan Turing Institute that tried to find evidence of how AI had helped with the pandemic, uncovered a range of issues. In a report published last year, the researchers identified widespread problems accessing health data needed to use AI without bias, according to a recent account in Wired.
“We wanted to highlight the shining stars that show how this very exciting technology has delivered,” stated Bilal Mateen, a physician and researcher who was an editor of the Turing report. “Unfortunately, we couldn’t find those shining stars; we found a lot of problems.”
Many ideas for the use of AI in healthcare have not progressed beyond initial proofs of concept, often because an adequate quantity and quality of data is not available to properly test. Companies working on facial recognition systems, for example, are able to access billions of photos to improve image-recognition algorithms, but accessing health data is more difficult because of privacy restrictions and uncooperative IT systems. Also, deploying an algorithm that will make recommendations on an individual’s medical care carries risk.
“We can’t take paradigms for developing AI tools that have worked in the consumer space and just port them over to the clinical space,” stated Visar Berisha, an associate professor at Arizona State University. He warned in a recent journal article that many healthcare AI studies make algorithms appear more accurate than they actually are, because they use powerful algorithms on datasets that are too small.
AI systems trained on too-small datasets can have blind spots. “The community fools [itself] into thinking we’re developing models that work much better than they actually do,” Berisha stated. “It furthers the AI hype.”
Stanford researchers found in a 2020 study that 71 percent of data used in studies that applied deep learning to US medical data, came from California, Massachusetts or New York, with little or no representation from the other 47 states. Low-income countries are barely represented at all in AI health care studies, according to authors of the Wired account.
The founders of the Nightingale Open Science non-profit are trying to improve the quality of datasets available to researchers. It is working with health systems to curate collections of data, including from patient records and medical images. The company attempts to anonymize the data.
“The core of the problem is that a researcher can do and say whatever they want in health data because no one can ever check their results,” stated Ziad Obermeyer, a Nightingale cofounder and associate professor at the University of California, Berkeley. “The data [is] locked up.
Digital Healthcare Platform Race Heating Up
The biggest companies in the world are powered by platforms, such as Facebook, Amazon and Google. In healthcare, platforms have been difficult to establish. A digital healthcare platform today can rapidly deploy digital capabilities using cloud services.
Examples include Merative, formed by the buyer of the healthcare analytics assets of IBM Watson Health, Francisco Partners. The company will combine Watson Health with other digital healthcare assets it has been assembling into a portfolio, including Availity, GoodRx and Zocdoc. The company said the new entity will serve a range of healthcare clients. “We appreciate IBM’s work in developing this business,” stated Ezra Perlman and Justin Chen of Francisco Partners, in a press release. The company is reported by Bloomberg to have paid some $1 billion for the IBM healthcare assets.
Other digital healthcare platforms from the big tech players include Microsoft Cloud for Healthcare, Amazon HealthLake, Google Cloud Healthcare Data Engine, Salesforce Health Cloud and SAS Health.
More digital healthcare platforms include: Vantage Health, Doctor on Demand, Practo (originating in India), and Peppy, promoting employee well-being and healthcare.
“Health care is beginning to understand the economic power of platforms, especially in a data-rich environment,” stated Randall Williams, MD, managing director of Digital Care Advisors consulting practice, in a recent account in MIT Sloan. “The industry is in an early phase of experimentation.”
The push for value-based care, the movement toward consumerism, the federal mandate for interoperability, and the expansion of virtual and home-based care models are trends driving the healthcare platform approach, with AI at the center.
The platform works when the incentives among participants are aligned. “It’s not just about what the patients want, but also what the clinicians want," stated Susan Woods, MD, founder and CEO of remote patient monitoring firm Generated Health.
For example, patients often want nutrition advice, but physicians receive limited training in this. With a platform in place, it’s easier for a physician to recommend a nutritionist or a vetted nutrition app, she said.
New data models are being developed in part because the government is requiring IT vendors to make a certified API “endpoint” available to customers by the end of 2022. The API would enable data sharing and exchange between different healthcare systems.
In this environment, Williams said, “Every health care organization should be thinking of itself as a digital platform. If you’re required to use APIs, that effectively makes you a platform.”
Read the source articles and information in VentureBeat, Wired and MIT Sloan.
(Write to the editor here.)