There is certainly no shortage of research on artificial intelligence, machine and deep learning algorithms these days. You will come across a number of announcements every week reporting on fascinating findings, new methods, and breakthrough results.
Despite the unquestionable buzz, it is challenging to pinpoint examples that actually found their way into everyday medical practice, that became part of “standard” healthcare.
In this article, we will introduce three areas we believe will be the forerunners of the A.I. revolution in medicine and will discuss four principles that help to determine whether you can expect a certain technology to become mainstream soonish – or not.
For the sake of understanding, let’s start with the three areas briefly: A.I. in skin checking apps, vocal biomarker analysis and cough analysis are technologies that – we think – are closest to becoming widely used in everyday healthcare. We’ll get back to them a bit later.
For a given technology to penetrate healthcare, it has to check the following four boxes:
1. Has to offer low-hanging fruits
To justify substantial investments, you need to see – at least the promise of – tangible results. This is especially true if we are spending taxpayers’ money, and thinking about implementing technologies on a nationwide scale.
In some cases calculating the benefits can be pretty abstract: you achieve x more efficiency by covering z more cases with the same amount of medical personnel, resulting in y amount of saving while providing better care for the patients.
Using A.I. algorithms to determine whether a skin lesion is worrisome and enhance the diagnostics capacities of dermatologists by shortlisting the most probable options based on patient-submitted images has benefits on many levels. Patients get a faster diagnosis, doctors can go through more cases, unnecessary medical appointments can be avoided, the process becomes more efficient, resulting in financial returns.
And of course, these algorithms are ever improving, and will most likely recognize the rarest skin conditions with 100% accuracy in 20 years – but they deliver now, and all along the way.
2. Needs to provide definitive answers to specific medical questions
At this stage, one-trick ponies stand a better chance than intricate projects with a wide focus – if they do that one trick well.
Your general practitioner could benefit from an application that can accurately determine whether the cough of the patient is the result of gastroesophageal reflux, a viral infection or pneumonia – but he has no use for a complex deep learning algorithm that can model all possible protein structures in the universe.
3. Can be efficiently used without years of special training
Any A.I.-backed technology that has the ambition to become mainstream in healthcare has to be easy to use – not adding to the burden of medical personnel, neither in lengthy training nor in significant investments. The more it resembles methods already in use, the more it relies on devices already in the practice – the better.
A smartphone – already in the pocket of the doctor – running an app that can detect the first signs of Alzheimer’s disease through vocal biomarker analysis can be used by almost anyone. There will always be research projects and initiatives that will be impossible without in-depth specialist knowledge and years of training – but those, while might be also crucially important in the long run – will never become mainstream.
4. It can be fairly easily regulated
No medical novelty can be implemented on a wide scale without being thoroughly regulated. But regulatory hurdles around healthcare A.I. algorithms are complicated enough, especially if we think about adaptive, constantly improving models – that are simply nightmarish from this point of view. How can you set the boundaries around something that might be completely different by tomorrow?
A.I.-based technologies and applications that are suitable for easy regulations stand a much better chance of infiltrating everyday practice – as these are the ones that actually can be regulated as of today.
The Medical Futurist Institute published a study in Nature Magazine on the state of artificial intelligence-based FDA-approved medical devices and algorithms, you can find the constantly updated online database here. It currently includes 79 entries, and interestingly enough, based on the information content of these clearances only, you would be not informed about the given technology being A.I.-based in about half of the cases.
Bet 1: Skin-checking apps
Skin checking apps are our first bets on the shortlist of possible A.I. pioneers penetrating healthcare. There are a number of existing applications, working very similarly. You take a picture with your smartphone of a suspicious skin lesion and submit it via the app. It first gets checked by an A.I. algorithm, providing a fast evaluation of whether it looks malignant, which will be followed by a conclusive diagnosis by a dermatologist.
If you wonder if these apps have real-life benefits, just click over to this recent experience of mine on how we found a high-risk birthmark on the arm of my 15-year-old niece, that was described as a “ticking time bomb” by my dermatologist and got removed since then.
Skin checking apps easily tick all four boxes above, address an area traditionally burdened by the shortage in the healthcare workforce, are very easy to use both for the patients and the physicians, answer definitive questions and are easy to regulate.
Bet 2: Vocal biomarkers
A “biological marker” or “biomarker” refers to medical signs which indicate the medical state observed from outside the patient. Vocal biomarkers are medical signs deducted from the features of your voice. The characteristics of your voice – or as medicine labels them, vocal biomarkers – reveal a lot about your health and help in detecting serious diseases and health risks.
The field of vocal biomarkers is developing steadily, with some very practical applications emerging recently. We especially like this 2021 study that introduces how automated, early Alzheimer’s detection is possible through the analysis of voice patterns. Again: plain benefits, definitive answers, ease of use and no major regulatory pitfalls.
Bet 3: Cough/breathing pattern analysis
Similarly to vocal biomarkers, cough analysis and respiratory breathing pattern analysis are based on cough/breathing audio patterns, which – given you have a large enough set of data to train the algorithm – also provide diagnostics opportunities to detect infections or chronic conditions.
These initiatives target remote analysis through mobile applications, evaluating the cough and/or breathing patterns of the patients. Although COVID, in general, seems much less terrifying these days than it looked two years ago – still, any solution that offers remote diagnosis for an infectious disease is a welcome addition to the doctors’ tools. And again, cough analysis is in line with the principles above, making it possible to become widely used.
Many more to come
These three examples are just the tip of the iceberg, there is an abundance of already existing technologies – algorithm-backed smart stethoscopes, A.I.-guided ultrasounds, sepsis detection software used in hospitals, and so on – not quite on the brink of becoming the new normal in healthcare, but showing how diagnostics and prevention possibilities in medicine are shifting.
As A.I. is getting widely accepted in the medical community and among patients, penetration through the walls of the ivory tower will become smoother.
The post How Will A.I. Penetrate Healthcare? Through Your Skin, Voice And Cough! appeared first on The Medical Futurist.