AI and the Canadian doctor

Naylor

In the deluge of recent presentations on artificial intelligence (AI) and medicine,  one of the more relevant for Canadian doctors came at last month’s annual meeting of the Canadian Medical Protective Association (@CMPAMembers).

The video of that presentation has just become available and gives people an opportunity to hear from CMPA Executive Director and CEO Dr. Hartley Stern and Dr. David Naylor (@CDavidNaylor), professor of medicine and president emeritus of the University of Toronto and one of the most respected voices in Canadian health care.

Given that the CMPA provides legal advice to physicians, Dr. Stern’s comments were particularly relevant, and Dr. Naylor had the opportunity to expand on remarks he published last year in the Journal of the American Medical Association on the topic.

“Every area with images at its core for practice will become a realm where AI will have transformative effects ,” Dr. Nayor said, because of ability of a computer algorithm to quickly analyze individual pixels rather than just patterns and to characterize in much greater detail.

“Used intelligently (and) used to augment human intelligence, AI can streamline workflow and relieve us of drudgery and give us time to better physicians caring for people.” In addition, he said, AI can help solve complex analytic problems and capitalize of the widening availability and richness of data.

“Many of these algorithms will be at their best supporting us and not making decisions on their own,” he said.

On the negative side, Dr. Naylor said, AI has the potential to devalue judgement and dehumanize care, is dependent on the quality of information on which algorithms are based and makes decisions where causal pathways are hard to determine.

When it comes to the potential uses for AI in medicine, Dr. Naylor said, “these are early days,” noting Dr. Eric Topol’s assessment that few of the algorithms currently in use have been rigorously tested and evaluated. And, he said, nobody has yet established how to critically evaluate publications dealing with AI because they vary so much.

Dr. Naylor said physicians should take the middle ground and not be stampeded into either unquestionably rejecting or accepting the value of AI in medicine. The integration of AI into medicine will either be smooth or disruptive, Dr. Naylor concluded, and which way it goes will depend to a large degree on physicians.

Speaking after Dr. Naylor, Dr. Stern reiterated the potential benefits and challenges for using AI and deep learning in medicine. “There is great promise that we can improve diagnostic accuracy,” he said, and  the ability of physicians to improve treatment plans can be improved, while reducing costs and the overuse of medical tests.

The promise, if properly implemented, is that AI will be able to transform the healthcare system and in so doing improve well-being and quality of life for physicians, Dr. Stern said.

However, he added, the regulatory framework and legal environment for using AI lags significantly behind the development of the technology.

For AI to be properly integrated into health care, Dr. Stern, said individual physicians need to be able to trust that the technology will do what it says it is going to do. One of the roles of CMPA, he said, will be to provide a bridge between the interest in these technologies from physicians and patients and their trust in them.

With poor communication being a main reason for complaints about physicians to regulatory authorities and the CMPA, the ability of physicians to properly explain the algorithms used by AI to their patients will be of critical importance. “You are going to have to learn how to tell that patient what this AI is going to do for you.”

Dr. Stern went on to then elaborate a number of key challenges to integrating AI including considerations of patient privacy while gathering the immense amounts of data required to develop reliable algorithms.

“In our environment, health care professionals are accountable clinical diagnosis and treatment plans. It’s you, not the machine.” He noted the Canadian regulatory and legal framework is not yet established for determining accountability when a wrong diagnosis is made based on an incorrect algorithm.

Dr. Stern urged the audience to get involved with medical associations and regulatory authorities in developing frameworks to provide adequate protections for physicians. “Without a clear policy, you are at risk.”

The session was accompanied by the release of a CMPA background document: “Can I get an (artificial) second opinion?” That document notes: “While AI can provide information for you to consider, it is important to ensure that actual medical care provided to the patient reflects your own recommendations based on objective evidence and sound medical judgment.”

(Image: Dr. David Naylor from CMPA video)

Morphing with Moomins: #HIMSSEurope19

Moomin

To be honest, one of the main reasons to write this blog was the chance to use the headline – which probably needs some explanation. And if the explanation doesn’t make too much sense then I can only blame jet-lag and a degree of digital health overload from attending the first day of the HIMSS Europe 2019 digital health conference in Helsinki, Finland.

And just to be even more inexplicable, the main theme of this blog is – as it has been a few times this year – AI and its applications in health care.

It is currently impossible to avoid AI as a conference theme/main topic of conversation at health technology or digital health conferences. In fact, as this conference goes on in Helsinki another conference in Boston dedicated specifically to AI and machine learning is just about to begin. To quote from an AI session description here at HIMSS: “Personalized medicine, predictive analytics, augmented diagnostics, real-time processing of wearable data… AI is buzzing and feeding our imaginations with hopes and dreams of a truly data-driven healthcare with optimal clinical and financial outcomes.”

But what struck me while listening to that well-attended sessions is that AI has morphed from Big Data as a major theme in digital health care.

Remember how just a few years ago everybody was talking about Big Data (with the capitals) and the potential benefits for care delivery from the incredible amount of data being accumulated on all aspects of people’s health? Well, it is clear that this theme has been coupled with machine learning and artificial intelligence to take the potentiality to another level. Now when speakers – such as those from the Helsinki University Hospital (HUS) talk about the “data lake” they are filling with a wide range of genetic, demographic and health data from their patients, they are doing it in the context of taking all of this information to create algorithms to enable AI use.

In fact, according to session moderator Dr. Mikka Korja, a neurosurgeon at HUS, Helsinki is one of the world leaders in exploring the use of AI in health care with 100 projects underway here.

That data are critical to effectively utilizing AI in health care is without question. But, as speakers repeatedly noted, it is not just the quantity of data points needed to create effective algorithms but also the quality of data required. It is clear that effective use of AI is impossible without standardized digitized data collected in an electronic format. The point was also made that most health care organizations are currently not configured to continually gather data and feed it back into the system in a manner that is required to develop AI functionality.

Most of the speakers at the HIMSS session discussed the potential applications for AI to improve clinical diagnosis and this remains the most sexy aspect of AI. Yet, I thought it was actually Dr. Kaveh Safavi, senior managing director at Accenture Health, who gave the most compelling presentation when he noted that enhanced diagnosis is actually one of the more challenging roles for AI to play in health care. During question period he noted that that for AI to augment diagnosis requires a high level of machine learning with masses of data which might require years of work and with all of this compounded by dealing with privacy and security concerns.

Where AI can and will have a huge impact, Dr. Safavi said, is in cost reduction and improving the consumer experience, in addition to outcome improvement. He argued that effective use of AI will help reduce staffing requirements and increase capacity at a time when “we’re running out of people to provide health care.”

Dr. Safavi also quoted an Accenture study showing the main benefits from AI in health care will come in areas unrelated to diagnosis with the biggest estimated potential benefits being in the areas of; robot-assisted surgery, virtual nursing assistants, administrative workflow assistance, and fraud detection.

For all the hype you hear about using AI to better detect melanoma or leprosy, Dr. Safavi’s comments are worth serious consideration.

Oh and about that headline. For those unfamiliar with Finnish culture, Moomins are an incredibly popular family of fictional creatures created by Tove Jansson in the 1940s and featured in books, comic books, TV, plush toys etc. And at a conference in Finland who could resist working them into a headline.

 

 

Radiology and AI: An algorithm took my job

Radiology18

Radiologists and those who love them sometimes seem to be some of the most insecure people around and radiology seems consistently voted the specialty most likely to be made defunct by modern technology.

In Canada, this perception has been voiced repeatedly over the past decade or two, first with the perceived threat of radiologists on the other side of the world working during our night to interpret scans and provide results when Canadian radiologists are sleeping, and now with the widespread attention given to artificial intelligence (AI) and deep learning and its potential uses in health care.

At last year’s annual meeting of the Canadian Association of Radiologists (@CARadiologists) I pondered whether this year’s meeting of that organization would be leaving it too late to discuss the significant issues revolving around AI and radiology (see my post “(A) I, Radiologist”) given the speed with which the field is advancing.

I need not have worried.

Last week the CAR released a white paper on the role of AI in radiology in advance of its annual meeting this year to be held April 26-29 on the theme of AI in radiology with the learning objectives including: discussing the recent changes that have occurred in imaging as a result of the implementation of artificial intelligence, deep learning, and machine learning in imaging workflows and; discussing the opportunities of big data and artificial intelligence to improve on the diagnostic performance and predictive value of imaging.

As an extensive overview of the topic, the CAR document squarely confronts the main issue at hand, namely that “recent breakthroughs in image recognition introduced by deep learning techniques have been equated in the media with the imminent demise of radiologists.” The authors go on to state rightly that the work of radiologists goes far beyond that of just correctly interpreting images.

“… the complex work performed by radiologists includes many other tasks that require common sense and general intelligence for problem solving–tasks that cannot be achieved through AI. Understanding a case may require integration of medical concepts from different scientific fields (eg, anatomy, physiology, medical physics) and clinical specialties (eg, surgery, pathology, oncology) to provide plausible explanations for imaging findings. Such tasks accomplished by radiologists on a daily basis include consultation, protocoling, review of prior examinations, quality control, identification and dismissal of imaging artifacts, cancer staging, disease monitoring, interventional procedures for diagnostic or therapeutic purpose, reporting, management guidance, expertise in multidisciplinary discussions, and patient reassurance”

However, the white paper’s authors also fully appreciate the potential of AI and state that “to remain current Canadian radiologists will need to follow and contribute to health care AI research and development, embrace the changes in workflow that will be required to support the implementation of clinical AI and adapt to changes in their practice that will improve care of their patients.”

So, while the demise of radiology (again) does not seem imminent, the white paper will bear close reading by Canadian radiologists who wish to remain relevant with some of the most significant and fundamental advances in medicine currently underway.

(Image courtesy of the Canadian Association of Radiologists)