The last 6 months have seen an increase in FDA approvals and we can expect to see even more approvals. Verbiage used in some of these FDA approvals include ‘analyzing radio-logical images using machine learning algorithms to detect and diagnose fractures’,
‘Using artificial intelligence algorithms, the device is able to determine whether a patient has referable retinopathy’
and ‘device provides triage or notification that is informed by machine learning, artificial intelligence or other image analysis algorithms’. The verbiage is clearly encouraging as these tools offer another layer that can help clinicians with decision support.
Equally though there is also the fear that AI will replace healthcare providers, for example Silicon Valley-investor Vinod Khosla said that machines will substitute 80 percent of doctors in the future in a healthcare scene driven by entrepreneurs, not medical professionals, or when Professor Geoffrey Hinton, the godfather of neural networks said that it is ‘quite obvious that we should stop training radiologists as image perception algorithms are very soon going to be demonstrably better than humans’. In fact, in May 2017 the Professor of Radiology and Biomedical informatics at Stanford University, mentioned how he received an email from one of his students saying he was thinking about going into radiology but did not know whether it is a viable profession anymore
The U.S. Department of Health and Human Services (HHS), with support from the Robert Wood Johnson Foundation, asked JASON to consider how AI will shape the future of public health, community health, and health care delivery. JASON is an independent group of elite scientists which advises the United States government on matters of science and technology, mostly of a sensitive nature. JASON is a reference to Jason, the mythical Greek hero who was the leader of the Argonauts.
The report states that the design of future health care information systems revolve around two questions that need to be answered, one concerned with computer science and the other with fundamental biology. The computer science question is, whether an entire medical database can be created and used with the data maintained in a form accessible to human cognition, avoiding the cumbersome and costly translation from analog to digital. The fundamental biology question is whether the natural coding of information in a human brain is basically analog and not digital.
One known fact is a mathematical theorem proved by Marian Pour-El and Ian Richards in 1978. The theorem states that analog computing is in a precise mathematical sense more powerful than digital computing. Pour-El and Richards display a number that is computable with a simple analog device but not computable with any digital device as defined by Alan Turing in his famous paper, “On Computable Numbers’’ in 1937.
Their discovery gives us reason to hope that a new generation of computers operating as analog devices may give us databases more user-friendly to us than our present-day digital databases.
The second fact, supporting the view that the human brain operates as an analog device, is our subjective experience of perception and memory. We experience the visual operation of our brains as a rapid and effortless scanning of pictures moving in space and time. To our subjective view, the brain appears to be primarily a device for the direct comparison of images. We see the images as whole scenes with shape and style, not as collections of pixels. Our perception of continuously moving images does not prove that our brain is an analog device, but it makes this a plausible hypothesis.
The truth is there was already a lot of innovative focus on developing AI medical solutions (whether effective or not), but lacked support from effective approval processes. FDA approvals basically mean that you can get paid for services rendered. There has also been a large amount of investment for example Healthcare organizations surveyed by Optum said they planned to invest an average of $32.4 million over the next five years on artificial intelligence.
A number of factors are responsible: firstly the 21st Century Cures Act (Cures Act) was signed into law on December 13, 2016 and was designed to help accelerate medical product development and bring new innovations and advances to patients who need them faster and more efficiently. More specifically the Act specifically exempted certain software products from the definition of ‘medical device’.
Secondly, Scott Gottlieb became FDA commissioner in May 2017 and under his leadership the FDA has made fighting the crisis of opioid addiction a top priority, advanced initiatives on addressing drug pricing, banned the sale of most flavored e-cigarettes in tens of thousands of convenience stores and gas stations across the country and is expected to propose a ban on menthol in regular cigarettes.
He was named “50 People Transforming Healthcare in 2018” by Time magazine and ranked No 6 by Fortune Magazine's annual survey of the “The World’s 50 Greatest Leaders” in 2018. He also believes that machine learning can help advance healthcare.
‘One of the most promising digital health tools is Artificial Intelligence, particularly efforts that use machine learning.’ Scott Gottlieb, M.D, FDA Commissioner
There have also been other notable non-AI based advances from the FDA this year such as the approval of 45 novel drugs and biologics which has been the most approved in more than 20 years. Clearly, the FDA is encouraging an environment of innovation.
Equally important are the advances in the development of fast hardware Graphics Processor Units (GPUs) allowing for the training of much larger networks, the availability of large labeled datasets both of which have given rise to the ‘data-driven paradigm’ of Deep Learning (DL). Unlike previous eras of excitement over AI, the potential of AI advancement in health may make this era different because of the merging of the following three forces:
1) frustration with the legacy medical system, 2) ubiquity of
networked smart devices in our society, 3) acclimation to convenience and at-home services like those provided through Amazon.
Yes, this may be old news…but how does the US government describe AI? A report by the Executive Office of the President National Science and Technology Council Committee on Technology stated that “There is no single definition of AI that is universally accepted by practitioners. Some define AI loosely as a computerized system that exhibits behavior that is commonly thought of as requiring intelligence.” And will today’s artificial intelligence still be tomorrow’s artificial intelligence? As software becomes increasingly capable, tasks considered as requiring “intelligence” are often removed from the definition, for example, optical character recognition is usually excluded from ‘artificial intelligence’ has become routine technology. This phenomenon is known as the AI effect, which can be described using Tesler’s Theorem “AI is whatever hasn’t been done yet.”
The EU is also introducing new regulations that will apply as of May 26, 2020, regulations that contain additional provisions that specifically address software medical devices. Of particular relevance, software with a medical purpose of “prediction and prognosis” will fall within the scope of the Regulations. This means that AI software that currently is excluded from being regulated as software medical devices under the existing regulatory regime, because they do not provide a treatment recommendation, but only a prediction of risk to or predisposition of a disease, may in the future be reclassified as medical devices.
INVESTMENT: The European Commission outlines a $24 billion (€20 billion) investment between 2018 and 2020, with the expectation that those funds will come from public and private entities
On August 31, 2017, the State Food and Drug Administration (CFDA) released a new version of the “Medical Device Classification Catalog”, which came into effect on August 1, 2018. The original version of the catalog increased from 15 pages to more than 150 pages and the number of name examples increased six-fold to 6609, of which the proportion of medical imaging equipment increased significantly.
The CFDA has also added a specific category corresponding to artificial intelligence-assisted diagnosis, which is embodied in the catalog of analysis and processing of medical images and pathological images.
INVESTMENT: China is investing at least $7 billion through 2030, including $2 billion for a research park in Beijing. The Chinese government foresees a $150 billion AI industry at that time.
The UK government announced earlier this year that it plans to invest millions of pounds of government funding to develop AI that is able to diagnose cancer and chronic disease before symptoms have developed, potentially saving 20,000 lives each year.
Prime Minister Theresa May challenged health charities, the National Health Service (NHS) and the AI sector to pool data in order to transform the diagnosis of chronic diseases. In response to Theresa May’s challenge Jane Rendall, Managing Director of NHS imaging technology partner Sectra UK & Ireland, said: “The NHS has practically unused archives of millions of diagnostic images that could become one of the most powerful clinical data sets in the world if artificial intelligence is used effectively. Our health service has a wealth of imaging data that it can use to start teaching machines how to recognize parts of the human anatomy, and more importantly, how to recognize abnormalities.”
The UK government’s ‘Grand Challenge Policy Paper' published in May 2018 listed the following mission: Use data, Artificial Intelligence and innovation to transform the prevention, early diagnosis and treatment of chronic diseases by 2030.
INVESTMENT: UK announced a deal between private and public groups that would bring more than $200 million of AI investment into the country
In 2018 Canada introduced the ‘Regulatory Review of Drugs and Devices’ initiative, Health Canada established a new division within the Therapeutic Products Directorate’s Medical Devices Bureau to allow for a more targeted pre-market review of digital health technologies, to adapt to rapidly changing technologies in digital health, and to respond to fast innovation cycles. The initiative focuses on a number of key areas including AI.
INVESTMENT: The Canadian government has committed $125 million to AI research
The Food and Drug Administration (FDA) assures that patients and health care providers have timely and continued access to safe, effective, and high-quality medical devices.
A medical device is any apparatus, appliance, software, material, or other article, whether used alone or in combination, including the software intended by its manufacturer to be used specifically for diagnostic and/or therapeutic purposes
Medical devices are classified into 3 classes based on the risk, Class I, Class II, or Class III, with Class I being the lowest risk and Class III the highest risk. 3% of devices are are not yet classified by the FDA.
The FDA regulates over 190,000 different devices, which are manufactured by more than 18,000 firms in more than 21,000 medical device facilities worldwide.
The FDA’s 510(k) approval process dates from 1976, The 510(k) pathway is the most commonly used premarket review process. In 2017, the FDA cleared 3,173 devices through the pathway, or 82 percent of all devices cleared or approved.
Close to 20 percent of current 510(k) approvals are based on predicate devices that are more than 10 years old, a process that could be holding back medical device innovation. Under the 510(k) pathway, device manufacturers use comparative testing against predicate devices — devices already on the market — to show that a new device is as safe and effective as the predicate device.
The FDA introduced the De Novo classification for new, novel devices whose type has previously not been classified. It’s for devices that would otherwise be classified into Class III, and provides a means to classify into Class I or II.
The De Novo pathway provides a vehicle for establishing new predicates that can reflect modern standards for performance and safety and can serve as the basis for future clearances.
As a result, we expect to see more developers take advantage of the De Novo pathway for novel devices.
An Optum survey revealed that 91% of 500 US healthcare leaders said they expect to see a return on investment (ROI) for AI over the next few years. Hospital execs expect to see the ROI in four to five years, while health plans hope to see ROI in three years or less.
36% of respondents said they expect AI to improve patient experience, 33% think AI will decrease per-capita cost of care, and 31% believe AI will improve health outcomes.
“Analytics isn’t the end, it’s the beginning — it’s what you do with the insights to drive care improvement and reduce administrative waste,” said Optum Enterprise Analytics SVP and COO Steve Griffiths. “For AI to successfully solve healthcare’s biggest challenges, organizations need to employ a unique combination of curated data, analytics and healthcare expertise.”
Artificial intelligence in healthcare does present a whole new set of challenges around data privacy, security and ethics — challenges that are compounded by the fact that most algorithms need access to massive data sets for training and validation. These algorithms in some cases also ‘predict’ outcomes without the full knowledge of how the prediction was made. They are subject to bias and not fully transparent. This is leading to an emerging field of explainable AI aims to create new AI methods that are accountable to human reasoning.
Balancing the risks and rewards of AI in healthcare will require collaborative effort from technology developers, regulators, end-users and consumers.
Machine Learning Engineer | Health Informatics | Pharmacist | Mapping Medication.
Some of my other posts: