Daniel Harvey  

Hello and welcome to the Physiotherapy Research Foundation podcast. I am your host for today, Dr. Daniel Harvey. It's my honour to host as I past Physiotherapy Research Foundation grant recipient. I am currently the Director of the Master of Advanced Clinical Physiotherapy Program at the University of South Australia. I'm also a pain scientist and my work has frequently intersected with the digital world of innovation and technology and on occasion of artificial intelligence. This will be our topic for today and how these things intersect with the world of physiotherapy. 

Before I introduce my guests for today, we'd like to acknowledge the Turrbal and the Jagera peoples of the Meanjin as the original owners and custodians of the lands on which we meet, work and learn. We pay our respects to elders past, present and emerging. In the studio today, I have Professor Stephen McPhail and Dr. Ryan Gallagher. Perhaps rather than try and introduce themselves myself, I'll hand over perhaps to you first, Stephen, and just give yourself a brief introduction.

Steve McPhail

Hi, I'm Steve. Is that too brief.

Daniel 

Much too brief.

Steve 

I'm Steve McPhail, centre Director for the Australian Centre for Health Services Innovation in the Centre for Health Care Transformation at the Queensland University of Technology. I also Digital Health Informatics research at the Metro South Hospital and Health Service, which we've been on quite a long journey where we went from being fully analogue to fully digital and in the last. In fact I think we're the first large scale health service to be fully digital and paperless. So it's been an exciting journey. We've been on that journey for, you know, eight or nine years now and we're seeing the fruit of that. And along that way I became very interested in digital health and informatics and now increasingly artificial intelligence as we look to really translate some of this innovation into real world practice.

Daniel  

Amazing. Thanks. Over to you, Ryan.

Ryan Gallagher 

G'day i'm Ryan Gallagher I'm a I'm a physiotherapist first and foremost and have been for over 15 years now, but I am now a data science and data science manager at Honeysuckle Health, a digital health care and data science company. And I am also a conjoint senior lecturer at the University of Newcastle. After nearly 14 years of clinical practice in public health system, I learnt the power of data and how much that is going to transform health care. And the profession took the opportunity to retrain in data science and pick up some skills in that space. And been really been fortunate to find a space where I can apply. I apply my clinical knowledge and apply that apply the learnings in AI and machine learning to into a clinical environment to get some really, really exciting outcomes.

Daniel  

Cool actually, I'd like to start there if I can. Ryan What was it that made you as a clinical physio move into the world of tech and innovation?

Ryan  

Yeah, good question. I guess the first thing was doing a Ph.D. first and I did fall in love with the whole data aspect of nearly three years worth of data collection and then having that aha moment as you sat there and turned that into an insight and the realisation that, hang on, I'm no longer talking about helping a patient, I'm talking with talking about helping an entire population of patients and that I found that really powerful and fell in love with that and basically made a decision after doing my Ph.D. that I wanted to continue to do that I was going to go down the stats path because data science and stats and physio to IT basically in terms of professions. But I also realised that learning to code and having some skills in that space is going to be some really valuable moving forward. That's when I discovered. That's what data scientist do stats and coding. So that was the path that I went down and don't regret it at all.

Daniel  

Cool I know we were talking yesterday, I mentioned the idea of skill stacking where you, you take skills that often exist in one profession and you stack on skills from another. And I think they can be a real amplifies in terms of the impact that you can have.

Daniel  

Perhaps before we go a bit further, recognising that much of our audience might not be technophiles like we are, Could I throw it to use Steve and maybe you could give us a bit of background and even a definition of what is artificial intelligence, What do we mean when we talk about innovation?

Steve 

It's actually not that easy to answer on one hand, because many of the things that we call artificial intelligence, if you speak to some statisticians, they'll say that's just statistics. And of course, we're actually drawing on the same foundational base of of mathematics and computer science as well. And and applying these things together where we have advanced analytics capacity, which was sort of previously being held back by computational power. Now we're in this new era where we can do amazing things with data. And so some of the ideas that may have been good in theory we can now do in practice with our analytics and. 

So if you imagine we had, you know, very good statistics and then we supercharged it with, you know, computational power and coming and the lessons we went from computer science and put them together, we end up in this place of having artificial intelligence. It's not magic dust well, we do often get the feeling that, you know, if we just sprinkle artificial intelligence magic dust on things, it will it will somehow fix everything. It's not like that at all. We do need to be mindful in how we go about addressing our problems using artificial intelligence. But we do definitely have new opportunities that we didn't have before, and we're seeing new outcomes that we couldn't have achieved before without AI.

Daniel  

I think one thing that came across to me from your talk yesterday is that to some extent the label of artificial intelligence is a is a rebranding. And not to say it's not unique and more powerful than what we had before about what we previously might have thought of as statistics or a calculation or an algorithm, or maybe the next. The next thing was machine learning. Is it really just the complexity of the, say, the mathematics and the modelling behind it, combined with the extra computational power afforded to us by modern microchips? Is that what gets us from what we had before to now something we could call artificial intelligence?

Steve 

Oh yes, there are definitely key features. I think also we do often think of machine learning and artificial intelligence as the same thing. And artificial intelligence includes a number of other elements. But but machine learning is probably the one that we're seeing the most around when it comes to particularly when it comes to applied clinical practice. So at this stage yeah.

Daniel  

Yeah It's probably a nice segway. Maybe now we could move forwards and you could both share a little bit about how your work has intersected with technological innovation more broadly, perhaps, but in particular artificial intelligence. And perhaps I'll throw to you first there Ryan.

Ryan  

Yeah, Thank you. Yes. Echoing with what sort of Steve just sort of said that there's a bit of a misdemeanour in terms of some of the terminology and there is a bit of hype and a bit of laypeople labelling things as artificial intelligence when it's garden variety stats, particularly I hear in a like a scientific conference space, you've been hearing about machine learning for decades. You just haven't had a called that before. It's what most people call linear and logistic regression. What's been the really big buzz over the last ten years has been the ability to move beyond small datasets and move beyond Excel files and move to entire databases and to move to images and to move to using data in a format that hasn't been available to us before. So that's probably where a lot of the hype and a lot of excitement comes from.

Daniel  

Do you think that part of that is is also we're seeing it move from purely being in the research context now to being in implementation?  

Ryan  

Oh, 100%, 100%. And the pace in which it's coming along over particular over the last and last decade and now again in the last 18 months in terms of what's popped up. Yep. Is making it hit mainstream. People can see it firsthand and what it's doing and they're looking for opportunities to apply it into everyday use and naturally into clinical care. And if it's not the physios coming in with ideas for how to use it, it's patients coming in and asking questions off of things like Chat, GPT or Bing as it's now integrated into they're coming in asking questions off it. So we do need to be aware of what it is and understand and how to use it.

Daniel  

And what are some of those areas, particularly that you've been involved with?

Ryan  

The key things that I've sort of been looking at, I've got I've got a programme of work looking to use longitudinal data to actually in stroke outcomes. I've been looking at using MRIs of people's brains in the first 48, 72 hours after having a stroke and linking that with routinely collected rehabilitation data to build out an ability to identify can can we predict what someone's outcome outcome is going to look like after a stroke? We're still in a in the early stages of going through that work. But it is we are starting to see some promising results to be able to sort of predict. And it's probably a too strong a word estimate with more accuracy than what we probably can currently to understand that this is what it's going to look like for you as a stroke survivor or as a family member when you do leave hospital. So the ability to do that at scale with data that we wouldn't traditionally have used or looked at as physios before is quite interesting. Exciting. 

So I guess it's that combination of the the data that's now available to us in terms of radiology data routinely collected rehabilitation data. The ability to join those together link them, but then also having a computational ability to analyse that data, particularly the radiology data that we wouldn't have previously had even access to five, six years ago is super exciting.

Daniel  

I can see real uses for that in understanding prognosis and discharge planning for the future, identifying responders, potential responders to rehabilitation care. I can also say some real ethical issues there as well, because if you're trying to predict my, say, my mum's outcome after a stroke and then try to suggest her care should go one way or the other, on the basis of that, I know that I would feel quite uncomfortable about that.

Ryan  

Yeah, yeah. You raised some good, valid points. And I guess the key thing that's always going to come out of this is those ethical discussions and what this looks like in a clinical care context. I don't think you're ever going to have a situation where an algorithm is going to decide care for someone. I absolutely think you can have a situation where an algorithm is going to be no different to an MRI scan or to a blood test to a clinician in terms of it's a new layer of information to guide them that they wouldn't necessarily had. But. Just like a blood test doesn't diagnose you or an x ray doesn't diagnose you. A health professional does looking at that information. And I think that this will settle and these ethical conversations will settle on that same line, that you'll never have a situation when the algorithm says this. Therefore, that's what's going to happen. But you will have that new layer of information that is a new form of clinical test available to you to help guide your decision making.

Daniel  

Yeah, and I think something physiotherapy does really well, I think is have clinical reasoning at the core, not one particular test or one particular model of practice. So seeing it as just one more piece of information into our clinical reasoning processes seems to perhaps help overcome some of the potential ethical dilemmas that might come from it. In a moment, maybe I'll share a little bit of how my work has also intersected with innovation. And and AI first, perhaps I'll throw to you again, Steve, and you can share with us how your work has intersected with that space.

Steve  

Yeah, sure. Yeah. So really, the sort of work that we do, we want to bring innovation to life in health services, in practice, and make sure that that integration from research lab work progresses to clinical practice where we can impact the real world and to do that in ways that are sustainable and also well integrated with existing workflows and practices. Now, in the space of AI, that that creates a whole lot of new, I think of them as opportunities, but it's also problems as well. And so it's it has been a an interesting time to be alive when we have, I would estimate, thousands or maybe even tens of thousands of predictive models developed from AI algorithms of which very few, surprisingly few, have ever demonstrated an effect that of benefit in a real world clinical setting. And you're not going to hear this about we don't hear this with the hubris of Silicon Valley, you know, coming out. And when we see these amazing things with generative AI, Yeah, it's also exciting, but we can't forget that actually we need to check that we can convert what is a prediction into a meaningful change in practice that actually improves outcomes beyond what we were doing anyway. And so that's where the big gap is I think, at the moment. And so it's a it is an exciting time to be doing work in the in the implementation space as we as we look to move beyond algorithms and move beyond prediction. 

And even as Ryan was saying before, you know, we need to be able to ensure that when we link these datasets together and combine clinical information with a new layer of informatics that presented to clinicians in the right way so that can augment their decision making in a positive way, not just be another part of a noise interference in an otherwise busy lot of information get shoved at them anyway. And so that's that's where I think we're really making some good advances in how to do that. But it's a it's a long road. And I think the the advance of the lab work is progressing very rapidly and the advance in clinical practice day to day is still lagging behind, as it should to some extent, but perhaps lagging behind a bit more than than perhaps we would all like. There's a lot of reasons for that and some of the very good reasons. So we need to ensure we've got appropriate governance and accountability, clinical accountability. We've got to make sure things are regulated properly. And these might sound like really boring things, but they're actually super important because if we get that wrong, we lose trust. And once we lose trust and social licence to start to use advanced analytics in informatics to inform and augment our practice, then then we're we're going to be fighting a very uphill battle to ever win that trust back. So we've got to get it right from the outset. So how about you, Dan, tell us about your experiences.

Daniel  

Yeah, well, actually quite different to you guys. Artificial intelligence machine learning has intersected with my pain research in a couple of ways. I'm involved in a project led by Professor Tasha Stanton, where we had developing something analogous to mirror therapy for phantom limb pain. But in virtual reality, one of our approaches there is using muscle activity detected at the stump of the amputated limb, and we detect that muscle activity using EMG Electromyography and we use machine learning to train the software to understand what movement was intended by the person with phantom limb pain, such that we can translate that into movement of a digital limb. 

So I backup theoretical patient is wearing a virtual reality headset and they look down at where their missing limb was. But in place of that, we have, a let's call it a digital prosthetic and then we can make that look like it's a real limb or we can actually make it look like a prosthetic and then we can play games with them where they kick a soccer ball with their digital limb. But it's going to be different for everyone as to what particular pattern of muscle activity is going to relate to their intention to kick the ball versus bring their leg back again. So I guess that's an example of where we're using this technology to create new therapies. And a similar example of that relates to another thing that I'm working on, which is virtual walking for people with spinal cord injury. Related pain and people might not know, but something like two thirds of people have quite debilitating ongoing neuropathic pain after a spinal cord injury. 

And unlike something like mirror therapy, where you can put you're intact and moving limb on one side of the mirror and create an illusion that the affected side of your body on the other side is still moving. You can't do that if both sides of your body are paralysed and affected. But we can bring back the appearance that the body is moving in virtual reality and we can do that in the first person. And our approach to that so far has been to work with paraplegics who will move their arms and in sync to their arm movements. We have their legs moving, so they simply need to swing their arms while they're sitting in their wheelchair and and now find themselves walking through a forest again with real digital legs that are moving. And but what's turned out to be quite difficult is to make that feel natural. Intuitively, we thought I would just synchronise the right moving forwards with the left leg moving forwards, but if you sort of lock those together, you end up with something like a robot. 

And so now working with Zena Trost and Sylvia Gustin, who've created some software where they use machine learning to train a more natural gate. So it uses the sensor data from the headset and the hand controllers in order to learn how best to synchronise leg and arm movement so that it feels natural. I'll share one more example. Another broad thing I've been interested in, in the in the pain space is pain education and how changing beliefs can really impact people's psychological well-being. Pain itself and disability. But, you know, one of the things we've come up against in that space is the power of imaging findings can be very hard to convince someone it's safe to get back moving again if they have a very scary sounding imaging report. And I became interested in using. Actually, quite a while ago, I became interested in how we could rewrite imaging reports to make them honest, but less threatening to people. And my non-health professional cousin seems to have come up with a good, good idea. She had an ultrasound report recently on her shoulder, and it described all kinds of scary sounding things. And she was really quite anxious about it until her partner cut and pasted the imaging report into chat gpt with the prompt. Rewrite this and explain this in the voice of a pirate to give you some some idea of the power of that. Let me just read you a couple of sentences of what came out of that.

Steve  

Dan you've got to do this in a part voice. I believe it's the convention.

Daniel  

If I have to. Arrh ye see. In our ship's moving parts, there be a thing called the biceps. It's a tough a rope that helps us hoist the sails and left heavy loot. But alas, this biceps rope be getting swollen and makein' our matey wince in pain. The salty dogs call it biceps tendinitis. And the result of that was actually her being able to detach herself, at least from from that. And that seemed to have real a positive impact. You know, if that was my patient, I would have liked to have read it and screened it first, perhaps. But even without that, it seemed to have quite a profound impact. 

So as we get to towards the end of our time, perhaps it could be worth delving in a little bit more of what you see as some of the potential downsides and risks, maybe where you see that we might be overhyping this space. And if you feel to touch on that as well, where you think the future might be going from here, you're happy to take that one first, Ryan?

Ryan  

Yeah, absolutely. Yeah. I think there is a lot of hype in this space at the moment and is a lack of knowledge of what is actually going to look like. This is this is the modern day gold rush. And not getting caught up in that is going to be the key to it. I think the biggest risk is probably around inappropriate use and not understanding the limitations that comes with using AI. AI Is only as good as a data is provided, and that is the most critical success factor for any predictive algorithm, for any machine learning system is the quality of data. And if you are introducing biases into your data, then you are going to introduce biases into your algorithms as well. And it's critical that we understand that as health professionals, the implications that can come from that, not representing minorities correctly in our data, bringing in clinical assumptions into the data which may not necessarily hold true, will will have consequences for the accuracy of anything that we use. And then the issues still to be talked through and worked around in terms of data safety and data quality and what that looks like. And these aren't things that we're necessarily trained to understand as physios. 

We, we understand patient confidentiality and the need to maintain that. But what that looks like in a modern day world where you need that where the clinical information we collect is going to power the decision making tool we use in the future. We do have some work to do as a profession to build our understanding of what that looks like and probably recognising that our scope of practice is going to need to change in that space. And we're also going to have to expand our multidisciplinary teams to potentially talk about bringing in technologists and bringing in a new a new a new profession, new scope to help with this that may not currently exist.

Daniel  

Yeah, absolutely. I think because we're such a physically based profession, I think that it's not necessarily an inherent skill set within our profession at this point.

Ryan  

Yeah, Yeah. And I generally think it's going to need to be in the future because if we don't, someone else will. And that's the short of it. Patients are going to be coming to us more and more, wanting to use technology, wanting to understand how they can integrate it. And if we don't understand that as a profession and look to integrate that into our care pathways and into our training, we will get caught by other professions at do. But yeah, moving forward, what that looks like in terms of the traditional model of care and in terms of we're a very hands on profession, we're a very manual profession, integrating that with some of the work like what you're doing in terms of how do we take that from a physical environment to a virtual environment and using using technology and using AI to help with that. It's something that as a profession we probably need to start having more conversations at it, a bit more pace than what we are given how quickly this technology is progressing.

Daniel  

Thanks, Ryan. I'm sure a lot of that resonated with you, Steve.

Steve  

Yeah, Yeah, absolutely. I think that aside from the things Ryan just mentioned, which one's going to further depth but, you know, I can't we can't overstate the importance of getting the foundations right in terms of the data and many of the assumptions that we make about the quality of data that are entering systems that need to be verified, because often the false assumptions. But I won't go to the technical side right now. I think perhaps another consideration, particularly in the context of physiotherapy practice, is really about who is going to do it and how in terms of getting into practice. 

So if we're looking at products that are consumer products, some of these things are going to require TGA approval. We need to make sure we're aware of that. We can be buying products from overseas that don't even have TGA approval and then that has follow on implications for us. So need to be aware of when and when we when we do and when we don't need that. If we're trying to do things ourselves, we end up in a similar, similar sort of considerations regarding the regulation for safety. And you know, it sounds so boring saying all this stuff and the, the innovator in me wants to just, you know, wack myself over the head. But it is actually really important that we get, get, get that bit right. And then as part of that has how's it going to be funded right because this is yeah, it's like virtual care to some extent virtual care. There's some technology barriers providing care in a virtual context. 

But when it comes to maintaining algorithms, when it comes to integrating AI with our other systems from which we want to input data, particularly say, in a hospital setting or or in a large organisation beyond an individual private practice, there may be some supports for that if we're doing it in a practice setting, well, that's you know, there's going to be a limited technology budget and even in in large health services, there's a limited technology budget. And so when when the, you know, decisions need to be made about, you know, do we bring in this new technology which is going to have a flow on effect to 20 other systems in terms of connection and cyber implications? Privacy? Yeah, this is a real consideration. All that costs money to fund. And so as we look at our existing funding structures, I should also mention I'm a health economist. 

Maybe I should mention that I'll be very unpopular about the. Yeah, the idea of having new technologies come in and us committing time and resources to that, We need to actually demonstrate that it's worthwhile. Is it really worthwhile for our patients? Is a worthwhile for society in general? And I think that's where we need to commit some time and effort as researchers is to demonstrating that value. Because when we can demonstrate that the value is there for the patients and for society, then the the mechanism for funding just becomes almost a, you know, a financial engineering problem, which is much easier to solve than understanding or generating information to understand whether it really does have the benefits that we hope for, the cost it's going to take us. And cost is this cost in in cash cost, but there's also cost in our time and cost in patients time. And we need to understand that well so that we make good decisions and provide high value care.

Daniel 

I think that is a perfect note to finish on. So thank you both, Ryan and Steve for a really interesting discussion.

Ryan 

Thank you.

Steve 

Thanks Dan.