Medical AI could be a powerful force in helping to promote a radical democratization in health care.
Democracy, it has been said, is two wolves and a lamb voting on what to have for dinner. We may want the final say in politics, but the powerful can still dominate, to the detriment of the vulnerable. Democracy can be a double-edged sword.
Here, I argue that medical AI could be a powerful force in helping to promote a radical democratization in health care. We should understand the risks, but democratization in health care could also mean great benefit for patients. This concept may clarify the nature of the opportunity for medical AI companies.
Before considering machine-based medicine, we should consider human medicine. Modern medicine has delivered many spectacular successes. But we also need to understand that there is a big problem here. Medical error was in 2016 reported to be the third leading cause of death in the US; related to over 400,000 deaths a year in the U.S. alone.
This shocking statistic was the conclusion of a 2016 study in the British Medical Journal 1. ‘Medical error’ includes, for example, incorrect diagnosis, side effects of prescribed drugs and surgery-related infections. To put this problem into perspective: In the U.S., the yearly death toll from medical error is similar to the yearly death toll from Covid over the last two years. But these deaths from medical error occur every year.
How the Internet Helped to Democratize Health care
Before we turn to medical AI, let’s remember how the emergence of the internet led to a radical, and perhaps under-recognized, change in how medicine is practiced. It is now 27 years since the introduction of the Netscape browser, which marked the arrival of the wide availability of the internet.
Quite suddenly, far more medical information was available to ‘lay people’. They could now easily and meaningfully educate themselves on an unlimited range of health issues, even rare conditions. It became much more feasible for people to self-diagnose and even to have informed opinions on treatment options.
People being more directly involved in their own health care is sometimes called ‘patient empowerment’ and it can save people money, time and unnecessary suffering. Accordingly, it has been welcomed by the World Health Organization and by health services. Of course, people without medical training self-diagnosing and selecting their own treatment can entail serious risks.
People can be overconfident in their own abilities and make catastrophic errors. One vivid blog post hosted by the British Medical Journal stated "Do you remember that really irritating patient that came to ED (emergency department) at 3 a.m. with a 5-month history of neck pain, saying that the web told her that she might have a subarachnoid bleed? You probably thought – who ... are you to self-diagnose – I am the doctor, not you! …How do we balance the current philosophy of empowering patients … when they come to us spouting nonsensical rubbish?"2 Not all doctors support patient empowerment.
Machine Learning and Health care
Let’s be clear what about what is meant by ‘medical AI’. Most commonly, this refers to a branch of AI called ‘machine learning’. The ‘machine’ here is simply a software model trained on large datasets, using types of algorithms such as ‘neural networks’, ‘random forests’ or various ‘boosting algorithms’.
These types of algorithms work on different principles, but all identify patterns in complex data with minimal human intervention. Machine learning systems are unlike conventional ‘expert systems’, where algorithms are hand-crafted by the programmer. With machine learning systems, the programmer probably has very little knowledge of how the model works.
A neural network may have many thousands of ‘neurons’. Their ‘weights’ (how they turn inputs into outputs) can be inspected, but the model will be far too complex to be interpretable by a human. Even to the programmer, the model is a ‘black box’.
Machine learning has a huge range of potential applications in health care, in fundamental areas such as diagnosis, risk assessment and treatment selection. Also, medical AI has the potential to deliver the much-awaited but little-delivered promise of ‘precision medicine’. This is where the best treatment is selected based on patient’s individual characteristics.
The concept is that many relevant sources of data could be efficiently used, including not just symptoms, but also family history, genetics, blood biochemistry, data from wearables and many other potential sources of data. When advanced AI-based systems for precision medicine become commonplace, the practice of ‘one-size-fits-all’ medicine will seem primitive.
There is understandable resistance to medical AI from some quarters of the medical profession. Geoffrey Hinton, a pioneer in neural networks, has described radiologists as like Wile E. Coyote ‘They are already over the edge of the cliff, but they haven’t yet looked down yet. There’s no ground underneath. It’s just completely obvious that in five years deep learning is going to do better than radiologists.’ Hinton said that in 2016 and in 2022, plentiful scientific literature has already emerged demonstrating how machine learning systems outperform human experts in a range of medical fields, including many areas of radiology.
Medical AI can also be a force for democratizing health care and patient empowerment. This is because the tools of medical AI could be put directly into the hands of patients. Let’s consider two very different examples that illustrate this point.
Case study 1: Covid Diagnosis
Let us consider an example of how medical AI can promote the democratization of medicine, using the most topical area of medicine: Covid. Recently an Israeli team at Tel Aviv University led by Professor Noam Shomron, published a paper in Nature reporting that they developed a machine learning system (based on a ‘gradient boosting’ algorithm) able to predict Covid infection with 86% accuracy4. One of the beautiful things about this project was that the input data they used were so simple and easy to use in the real world. They used only eight binary features as inputs, listed below:
The potential benefits of rolling out such an AI-based system for Covid diagnosis could be substantial. Such systems could help Covid responses be more intelligently targeted, especially in countries with poor infrastructure for conventional testing. This could result in fewer infections, deaths, and restrictions on people’s freedom.
Importantly, such systems could also help address the huge environmental cost imposed by governments’ response to Covid, recently highlighted by the WHO. The global mountain of pandemic-related waste is not only about test kits; it is also about PPE waste used in face-to-face testing. There is no plastic waste involved in web-based diagnostics.
A simple web application could be developed at low cost and made available for free or at very low cost to users. Sixty percent of the global population now have access to the web. Therefore, it is clear how medical AI can be a force for empowering patients.
Case study 2: Home Ultrasound
At a recent investor relations event, ultrasound manufacturer Butterfly Inc., described its ambition to promote the use of ultrasound equipment in the home. Butterfly produce ultrasound devices so small they can fit in your pocket. The probe is attached to your phone, which displays the image. Another ultrasound company, Caption Health, has developed AI-based tools that can guide an untrained patient to get the probe into the correct position.
It seems like a short step from these kinds of technologies to home ultrasound devices that can help diagnose and monitor a wide range of conditions. Home ultrasound may become as common as home blood pressure monitors. To quote a recent review of AI in cardiovascular imaging In the long term, AI will not only assist doctors, it has the potential to significantly improve access to health and well-being data for patients and their caretakers. This empowers patients5.
What does this mean for the health care industry?
The potential for AI to improve medicine is enormous. As with most areas of human-machine competition, machines have the advantage that they are not constrained by hard biological facts, such as brain size. Machine-based applications can be trained on unlimited amounts of data in highly sophisticated ways.
They will continue to improve in terms of their accuracy and usability. Omdia’s research forecasts that the rapid rise in medical AI will continue in coming years.
Global projections of medical AI software revenue by use, 2019–26
Are there risks of the proliferation of medical AI? Yes, certainly and we will examine these in a future article. But putting these risks aside for now, it is now easy to envisage a world of AI-based home medicine. Based on the current level of technology, many, or even most, non-emergency conditions will be diagnosed at home using AI-based software, and AI-guided hardware.
After a diagnosis is received, a prognosis will follow and a range of treatment options, tailored to your personal symptoms, medical history and even genetics or blood chemistry. These systems may also list the probabilities, for you, of various benefits and risks of the different options. Such systems will give instant results, they will be cheap and very often be more accurate than the best human experts.
All this implies a fundamental disruption to health care. But then ‘fundamental disruption’ is what we should expect with the AI revolution.
What does all this mean for health care companies? There are huge opportunities, because new markets are emerging and there is scope for a greatly expanded customer base. Companies will need to remain agile, up-to-date with regulatory developments and constantly aware of their competitors. In Omdia’s forthcoming Medical AI Intelligence Service, this is precisely what we will cover.
Felix Beacher heads Omdia’s health care technology team. He has direct responsibility for the ultrasound intelligence service and is currently working on Omdia’s forthcoming intelligence service on medical AI.
1. Makary and Daniel (2016). Medical error—the third leading cause of death in the US. BMJ ; 353
2. On how the Internet changed medicine in the 21st century https://blogs.bmj.com/emj/2012/03/09/on-how-the-internet-changed-medicine-in-the-21st-century/
3. Longoni et al (2019). Resistance to Medical Artificial Intelligence. Journal of Consumer Research, 46, 4, 629–650
4. Zoabi et al 2021 Machine learning-based prediction of COVID-19 diagnosis based on symptoms. Digital Medicine 4: 3
5. Siegersma et al (2019) Artificial intelligence in cardiovascular imaging: state of the art and implications for the imaging cardiologist. Netherlands Heart Journal 27, 403–413