SICOT-PIONEER – Digital and Personalised Orthopaedics – 26th Jan 2024

🔰 SICOT-PIONEER

🔆Digital and Personalised Orthopaedics

🗓️ Date & Time – Friday 26 January 2024, 5:30PM

💻 Click here to Watch : https://tinyurl.com/OrthoTV-SICOT-72

🔯Moderators

KARTIK LOGISHETTY

SURESHAN SIVANANTHAN

❇Faculty

🔺BENEDIKT BRAUN – Germany

🔺RICHARD VAN ARKEL – UK

🔺CODY WYLES – USA

🔺BERND GRIMM – Luxembourg

✡TALKS

▪️Visualising fracture healing in real time: first surgical experience with the AO Fracture Monitor – Benedikt Braun

▪️Intelligent and personalised orthopaedic implants – Richard van Arkel

▪️Artificial intelligence: imaging, arthroplasty and beyond – Cody Wyles

▪️Wearables: digital mobility biomarkers in clinic and research – Bernd Grimm

👨‍⚕️ SICOT Education Academic Chair : GOWREESON THEVENDRAN

👨‍⚕️ SICOT President Elect : VIKAS KHANDUJA

👨‍⚕️ SICOT President : PHILIPPE HERNIGOU

🤝OrthoTV Team: Dr Ashok Shyam, Dr Neeraj Bijlani

📺 Streaming live on OrthoTV
www.orthotvonline.com

▶️ Join OrthoTV – https://linktr.ee/OrthoTV

OrthoTV : Orthopaedic Surgery & Rehabilitation Video & Webinars
One Stop for Orthopaedic Video Lectures & Surgeries

Subscribe for more videos: https://www.youtube.com/c/orthoTV
Register with www.orthotvonline.com for Exclusive videos
Join us on OrthoTV Social Media Cell : Remain updated about all national and international activities of OrthoTV:
*Instagram: https://www.instagram.com/ortho_tv_online/
*Twitter: https://twitter.com/OrthoTVonline
*Facebook: https://www.facebook.com/groups/indian.ortho/
*Linkedin: https://www.linkedin.com/in/orthotv/
*You tube: https://www.youtube.com/c/orthoTV
*Telegram: https://t.me/joinchat/EG43XkNyOQFvUF5iPek97g

OrthoTV: Portal for Orthopaedic Videos from around the globe

[Laughter] Right a very good morning good afternoon and good evening ladies and gents I’m vas kandja consultant orthopedic surgeon in Cambridge UK and also one of the founders of cot Pia and the president elect of seot and it’s indeed a pleasure to Welcome All of You logging in from

Various corners of the world into the webinar now in this Co world of non-con domains we’ve certainly embraced digital competitiveness at cot with the launch of cot Pioneer and through this platform we’ve done over 74 events and I’ve had over 100,000 views from 110 countries around the world so big thank you for

Joining us and following us as well now today’s webinar is the first webinar from the computers and enabling techn techologies committee and I’m ever so grateful to Kik bishe from Imperial London who’s the chair of the committee and also sures sanatan our Treasurer from Malaysia the vice chair of the

Committee along with the Stella faculty from all around the globe who’ve given up their precious time for this webinar and I’ll let karic and sea introduce them to you in a minute we’ll try and make this as interactive as possible so please do Post in your questions and if

You can’t join us today then you’ll access to this webinar on our ond demand platform once again a big thank you to all of you and hope you enjoy the webinar over to you goic can thank you vkas our first speaker is my dear friend and colleague Benedict

Braun who is an orthopedic surgeon at the University of tuban and he is also the chair of the smart digital Solutions task force from the AO he has done a lot of work on fracture monitoring and without further ado I’d like to ask Dr Brown to start his talk thank you very

Much so thank you sesh for the kind introduction and thanks for having me to share our first surgic surgical experience with the AO fracture monitor um these are just my uh disclosures keep in mind I am an AO person I do share a task force at the AO foundation and I am

The PCI for the fror monitor going to Market study um what do I want to talk to you today about so first of all what is the fracture monitor then I’ll give you a brief clinical background on the system and then uh I will share what I was invited to share the clinical

Experience that we had putting in the first two systems so the AO fracture monitor is basically on a very top level is a strain gauge with a battery and a Bluetooth sensor housed in a little uh metal housing that can be screwed on to various plates currently only uh femal

Plates so it’s large frag plates uh by DPS and the 41 medical by phasic plate and this sensor can record the implant load quasi continuously and transfer it via the phone um for up to six months without needing much of let’s say patient um patient compliance because it

Is an automatic transfer and does not require any any real interaction it um shares the load metrics with your physician and uh allows for daily daily update on on your weightbearing behavior and on the load that the implant is experiencing so this is just a just a

Top level overview of one of the earliest clinical studies that have been run with the AO fracture monitor it was in an tibial fracture revision cases and for this study it was mounted to an to an XIX or to a ring xfix system and basically during the healing time

Measuring the relative implant load and what you would expect in a healing fracture is that the load the implant experien is gradually decreases over time as a function of the heeled bone that is taking over the uh weight bearing of the patient um and this is the current status so now they’ve

Developed this implant to the point where it can be implanted actually and be screwed onto the plates that I just showed you and um the first clinical investigation it’s called a safety study but really is a going to Market study in Germany has started last October uh at

Four study centers so it’s between the university in hurg um and the uh University in mster and our Center which is the Big E hospital and University Hospital here in tubingen so now since I was expected to share our experience I will do exactly that so this is the very first patient

That was treated with this system as an implantable system that was done at our Center so it’s a 65y old male uh falling from a uh ladder and um he was initially treated with an XIX at an external institution and then we did our planning so metaphysical comminuted and joined

Rather simple fracture and we felt that would be a good uh indication to test the fracture monitor one of the specifics and yes we are pretty big on double plating is for this study since we do not want to take away any strain from the sensor that it is a fracture

That’s amenable to a single lock lateral plating system so just be mindful of this so um we took him to the OA and I’ll just walk you through the OA really trying to share the experience this is what the sensor look like it’s it’s just a regular implant housing and then it’s

Attached to the plate um basically if you if you’re familiar with the Locking attachment plate by by depu synes you can attach this uh this monitor because it works exactly the same way uh it’s got a screw and screw that’s put in with the large torque driver and then you

Mount the implant with a small Frack uh torque this is what it looks like when it’s mounted to the plate and uh you can already see that it’s it can be placed either that it’s sitting posteriorly or anteriorly on the plate and um anyone I’ve seen it uh uh done up to now

Intuitively we really placed it posteriorly on the plate because it just feels a little less imposing on the soft tissues for the patient and then um there’s one minor test involved so you do need to train your o staff um just to to be able to test the implant that is

Really transmitting um if you’re wondering how about battery life and and shelf life um within the casing there’s a little magnet and only when you pull it out of the uh out of the box does it really start uh the the the measuring process so the battery run time of half

A year is is um consistent in between the systems due to that safety mechanism then you test that it’s monitoring and then you can basically slide it in did a little larger approach since we had to look anteriorly into the joint but there you can see how it’s how it’s sitting in

There we just slid it in this is our team giving the thumbs up when we were done with the surgery um this is the intra op so we really did a um more or less Meo and try to not touch the fracture at all we were quite happy with

The with the reduction on the right hand side is the immediate posttop you can see in the latter how the system is sitting posteriorly on the plate not really impinging or anything so we we didn’t have any complaints on the patients yet and and during the surgery

Sliding in the plate you really don’t notice any difference to uh how you would put the plate in if you didn’t have the the sensor attached so this is the first patient we did back in October um so we do have some follow-up for this

Is the 8we um X-Ray when he was already in full weight bearing and you can see some some nice CS with with the with the medial cortical shell and also posteriorly this is the 3 months x-ray that was done a couple of weeks ago and

Um we we did go for a CT just to make sure and you can already see some briding K it’s not everywhere but it’s it’s progressing quite nicely and he is ambulating um and and getting ready to go back to work so he is doing quite

Well this is the second patient a little bit of a different case treated and external institution had a couple of surgeries 60-year-old female um initial straightforward injury had a lateral and a medial plate at the lateral plate taken out in a misguided attempt to to provide her with uh uh with a they

Called it a dynamization this is how she was introduced to our department broken medial plate and quite a significant valus this is her long leg standing um and you can see she does have a little bit of a valgus on the con side as well but um not nearly as much

As she does on the now non-union side so what we did is in the first step we took the plate out and and um cultured her and did a cement spacer we always dye our spacers blue in order to be able to better differentiate them from the from

The regular patient bone at the second surgery and we took the spacer out dialed in her um her various vgas with the large fra director which you can see in the in the in the far left image and then put the plate on again put the

Fracture monitor in in this cage uh case since we wanted to have on the one hand have stability for the fracture and on the other hand have the right amount of interfragmentary motion we chose to put on the AO basic plate which is marketed by 41 medical

And um really uh to to achieve the result that we want and to be feel safe that we can put her into weight bearing and quite early full weight bearing um as tolerated and this plate already has a capacity to to screw on the sensor without the screw and screw system one

Downside of the system is that this was the longest plate that we could get and it’s it’s it we would have wanted it to have to be maybe one or two holes longer on the on the proximal end but we put a locking attachment plate in just to to

To get some overlap with the with the uh FAL shaft uh this is her postop and follow-ups she was treated five weeks ago so I actually called in with her last week she’s doing fine and and ambulating well but she she was not here for x-ray followup just just

Yet what’s the Outlook so overall in Germany we have there there’s been four cases done um for the completion of the study we’re looking at 37 uh patients the primary end point since this is a going to Market and safety study is the the SAE rate which needs to be below 10%

Um secondary endpoint is performance of the implant and potentially a earlier diagnosis of the healing onset of a fracture currently treating Physicians as well as patients that would normally have access to the implant load we are blinded towards the data until the healing course is completed just to be

Sure um for the EC that we’re not influencing our treat based on the sensor data just yet so the ultimate aim is uh Market approval that’s it for the moment for the users experience and now I’m uh looking forward for the discussion at the end of of our talks

Thanks thank you Dr Brown for a fascinating uh overview of the uh smart play uh we can have questions at the end and I’ll let carik introduce our next speaker yeah thank you very much Dr Brown that was fascinating so our next speaker again it’s it’s a privilege and

An honor to introduce my colleague and friend Dr Richard van arle uh Dr Van arle is a senior lecturer in the department of mechanical engineering at Imperial College London he applies engineering science to Advanced Orthopedic interventions and he investigates implantable sensors the impact of surgery on joint bone biomechanics implant function additive

Manufacturing or 3D printing and new methods for pre-clinical analysis uh today Dr vanak will be speaking to us about his ongoing research into the next generation of Orthopedic implants and sharing with us his thought on the integration of am and sensors into devices which I hope will result in a

Step change in how we practice Orthopedics and troma surgery so without further Ado Dr Richard van Aral thank you thank you ktic let me share my slides uh so I’ll be talking about intelligent and personalized orthopedic implants I do have disclosures particularly for the intelligent implants content I have IP and shares in

The spin out company uh I’ll split the talk into two we’ll do about twoth thirds on intelligent implants and about onethird on personalized implants and for intelligent implants these have actually been around for a really really long time so since the 1990s we saw load sensing implants going into hit there

Were pressure sensing alternative and this expanded through into knees and if we go as far as the 2010s in the research domain this line of work really culminated with the orolo database if it’s not something you’ve come across it’s it’s truly superb it’s all open access it’s all freely shared you can go

On to ortho.com and it’s fascinating hips knees shoulders spines uh you can see videos of patients doing daily activities from Aqua gym to picking asparagus to jogging and you can get a sense for the load that goes through the implants uh really really exciting and really amazing that they shared such a

Body of work completely open access for everyone to look at and see uh as we move into current time we see that this is moving from a research sphere to a commercial sphere and so Kary medical launched the first commercially available smart implants and we see the AO fracture monitor is under clinical

Trials and this is really really exciting because it’s opening a new pathway in Orthopedics uh the canary device is indicated for clinical decision making and it’s only indicated for patients who would otherwise receive a extension and so this leaves us a few questions as we look forwards for can these be used for

Routine use and I think there’s two key questions can we make them small enough and cheap enough that they could truly be Mass market and can they influence that clinical decision making process so with regards to small enough and cheap enough we played with this idea and just

Saying how little can we actually put into a line plant to make it intelligent and so this is a small $11 sensor uh it’s about a 1 cmet footprint and about 1 millimeter thick with a small wire coil which is about 10 millimet in diameter no batteries no onboard circuit

Board that’s just it so it’s very very simple really pushing it down to the bare minimum we could have for a wireless smart implant device and with just a single $1 sensor we can get two useful bits of information we can say is this implant well fixed so the black

Line is a well fixed implant the red line is a non-well fix implant and the bottom access is different users at different distances taking their readings so yes we can get information about fixation and we can also get information about the temperature of the joint with just a single tiny $1

Sensor so then the question is can you influence clinical decision making so this idea of looking at infixation you could see that could have some sort of influence so we could first look at how that embodies in even the smallest implant so this is a fixed lateral Oxford uni compartmental device you can

See the sensor really is Tiny then there’s an external coil that would wrap around the patient’s knee and we got different clinicians to interact with the device and see if they could take a reading in the lab and uniformly across the board we could see before the implant was

Implanted when it’s well fixed and across three different states of loosening partially fully or even when the implants displaced they all across so that could potentially influence decision making but really the question mark Falls between is it septic or aseptic loosening and so we’re also interested in can we directly measure

Implant infix implants infections so this is a separate device we have a functionalized electrode that’s effectively picking up things bacteria emit but the natural body emits in much lower Doses and most of this implant device is actually hardware for an alternative thing I’m not going to speak about today the circuitry required for

The implant section the implant infection sensing is Tiny and there is an onboard battery in this Wireless embodiment and to test a device like this you actually run into a barrier that uh classically you’d want to test it in an animal but particularly for things like infection uh it’s expensive

And the animal burden is really really high uh so for our first stage of testing we’ve actually developed a way we can screen these and try these much quicker in the lab uh We’ve developed a bioreactive system where we keep bone living in the lab and it’s really accelerated our development pathway so

This looks like a histology slide taken from a living animal we’ve got osteoblast osteoclast all the classic burn tone over signs but we’re doing this in a lab based model and that really helps you accelerate how quickly you can do this research because it’s much cheaper and our animal burden is

Much lower uh so this is uh taking a femal head from a sheet we bring it into the lab and we implant the device and we then cult them in a bioa system half were implanted completely sterile half are sterile but then we expose them to inoculation loop with staff EPI and all

The data I show you is being recorded wirelessly with a mobile phone and within just five hours we can pick up that bacterial infection in the stfip exposed devices but normal bone turnover normal biological processors aren’t reading as false positives on the sensor so that’s lab it’s never quite the same

Doing it lab as and doing it in a full animal trial so we do then go further to animal trial uh this is the same implant system with the other technology that I’m not talking on today in sheep in trial uh and we’re playing with the idea can we actively influence that treatment

Pathway and so in the yellow line we’ve actively intervened based on sensor data and the black line we haven’t and we do see differences in healing responses based on this driving of an intelligent implant so can we go small enough and cheap enough we can go really really

Small and really really cheap and could we generate usable data that could influence the clinical pathway yes we believe that technology is in the pipeline and coming so for the second part of the talk I will slightly fix Focus to personalized implants and the original intent here was to talk exactly as

Cartic ated uh through additive manufacturing and we’ve had a long program of research looking into additive manufacturing for Oro uh we started just studying the core machines we studied what it did well what it did badly taking it all the way through to Technologies where you could get screw

Level fixation from a pullout device or we could tune the material properties of an additive manufactured implant to match bone and we saw in animal trial you got faster bone growth into the device when you tuned it to the Bone and we’ve looked at how we could optimize

Hip implants to give more normal bone biomechanics but what’s super excited about this is it is becoming commercial and clinical so there are multiple manufacturers with this sort of Technology freely available well freely comes with the price but this technology is on the market and it is impacting

Care and I think that’s really exciting so instead I thought I’d do a little different slant on personalized medicine and it’s nice to highlight personalized implants can just be the use of the implant for that patient and we’ve had a program of research where we’ve been studying the different types

Of knee OA uh so whether they’ve got a varus knee and getting medial OA vus knes lateral o maybe their patella out or just plastic patella um and they end up with patella fem o or combinations of different disease Pathways depending on their knee these are very distinct different causes different impact for

The patient and different Pathways and there’s a large number and an even split across patients so we studied how do these Pathways influence across patients the purple is individual arthritis in one of those compartments the blue is a mix across two depart compartments and only the yellow the 17%

Or one in six only one in six people have total knee osteoarthritis but when it comes to intervention everybody is getting a total knee replacement so clearly for single compartment there are solutions available partial needes have been around for a long time but the question mark is what happens if they further

Degenerate or what happens if they present with B compartmental disease at the time of surgery what can go in this space and is it really possible to mix and match partial knees in the same knee and so this is where we’ve been focusing a lot of research in the last five years

And what we found is taking this compartmental approach it really the evidence Stacks up it’s it is a nice thing for patient biomechanics so we’ve measured better knee extension function better knee stability it’s better for bone and it was better for patient gate and better for patient fronts and so

It’s a building body of evidence just using existing implants and I look at things like robotic where partial needs classic being associated with a steeper learning curve and a harder operation to do so combining them is doubly hard and we look at the technology that’s available and it leaves me asking the question

Well could we just use this established technology to deliver more personalized surgery without having to wait the next five 10 years for these things to translate through for patients so three learning points intelligent implants can be really small and really low cost we’re building the technology to directly influence clinical decision

Making based of data you extract from the implant but in the meantime maybe we should start thinking about how we can use existing technology to be more personalized and more targeted for patients because it is all available no good research happens on its own thank you very much collaborators thank you

Very much for funding and of course thank you very much for the invitation to speak today thank you very much Dr Van alal so we’ve had two great talks about devices that are either implanted in patients or around patients or attached to existing devices uh we’re going to change gear

Next and uh we’re going to be speaking to Dr Cody WS so again it gives me great pleasure to introduce Cody he’s an assistant professor of orthopedic surgery and a hip and knee surgeon at the Mayo Clinic in Rochester Minnesota he’s also the director of the orthopedic surgery artificial intelligence Lab

Osale at at the Mayo Clinic today Dr WS is speaking about his ongoing work at oale where he and his team are developing AI algorithms to assess Orthopedic Imaging helping us to predict outcomes in patients and generate new understanding beyond what the human eye sees or traditional maths has been

Capable of in the past over to you thank you cark thank you for the very kind introduction and it’s a real honor and pleasure to be with this esteemed group uh this morning over here in the central United States and uh early afternoon where most of you are

At I had the great pleasure last week uh to host cartic over here at Mayo Clinic in Rochester Minnesota this was sort of uh basic reciprocity after carik was so uh kind to host me on the British hip Society traveling Fellowship along with Mr kandja and others on the call just

Over a year ago but it was a real highlight for all of us at Mayo Clinic so there is a picture of myself with cartic my uh senior mentors Rafael Sierra and Rob trusdale and we just had a wonderful time um learning from cartic as much as sharing what we do here at

Mayo so today I’m going to be talking about artificial intelligence as cik mentioned um and really focusing on some of the high impact areas as I see them with nii coming down the pike for Orthopedics we’re going to talk about patient phenotyping efforts surgical planning and surgical execution Tools in

Particular and some of the key developments that are driving our ability to tackle these are areas are evolution of Big Data the recent advancements in large language models and most exciting for me the realm of generative Imaging so beginning with patient phenotyping uh we’ve worked on a series

Of multimodal risk calculators for some of the most prevalent and important uh complications of hip replacement that’s dislocation and paraprosthetic fracture and we wanted these calculators to answer two big questions the first is to determine patient specific risk the second and I would arguably much more important task of these calculators

Was to determine the degree of risk modification that we can achieve with surgical decisions within our control in operative theater so to create these calculators we evaluated over 30,000 patients operated on over a 20-year period at Mayo Clinic we Define them based on a series of non-modifiable risk factors including demographics and comorbidities

And modifiable risk factors things like our implant choices and Surgical technique so we’ve uh published these you can look those up if you like but we’ve uh built a series of AI models that allow these to work on real patients in action so I’ll show you an

Example of this this is a patient of mine an 80-year-old female body mass index of 30 undergoing a primary hip replacement for osteoarthritis she has a history of osteoporosis and a mild cognitive impairment and a degenerative spine but no prior spine surgery so here’s a graphical user

Interface this is actually live in our um electronic medical record now can be used for any patient so you see me clicking in all of those metrics I just mentioned to you can upload a patient’s preoperative X-ray and what you’ll get is a matrix output for this patient sort

Of in heat map fashion to show you their risk for dislocation and you can see on the x- axis on the bottom here we have surgical approach on the y- AIS we have combinations of implants like large and small femoral heads different acetabular liners standard elevated dual Mobility

You can see in the upper right hand corner this paat potentially very high risk for dislocation nearly 9% I can get that risk to less than 3% with some decisions directly under my control here’s the corresponding paraprosthetic fracture risk Matrix for the same patient once again we see potentially quite high risk for

Paraprosthetic fracture uh our variables that have changed in this scenario on the Y AIS are whether we use a cemented or cementless fixation for the femoral component whether or not we use a collar and again you can can see that based on some simple choices within our control

We can move the needle quite drastically on this patient from a high risk for fracture to a very tolerable risk near 1% transitioning now to large language models these have gotten a lot of attention recently of course and people focus on them talking about their Adept adaptability as physician co-pilots

Their ability to potentially decrease administrative burden provide smart care recommendations and I think all of those are coming all of those will be extremely important but one of the areas I’m focused on with this is automated registry curation so as uh has been well advertised in the media large language

Models are all the rage now but they’re so incredibly recent I think we forget this so the Arc of progress is incredibly steep at this point in time the first one was uh just on the scene in late 2019 and you can see the rapid spin-offs that have culminated in GPT 4

Not too long ago so large language models it’s important to kind of differentiate them in a binary fashion into the commercial models like gp4 and the open source models like meta’s llam chat um with commercial models they’re trained on more data they’re potentially more what we might call intelligent they have a

Lower bar for entry but they do need data deidentification which is the big uh barrier for Health Care use the open- source models they do require a little more tweaking you have to run them on a local infrastructure but that data does not leave an institution so it provides an

Opportunity to keep it all within your firewall and not have to worry about patient privacy concerns we’ve been trying to tackle this with some deidentification uh algorithms where we can take an original operative note like you see on the left run it through a Bert based the

Identifier and get some sort of dummy outputs that we can trace back on the back end for us to De identify these that’s one potential solution to use the more powerful chat GPT type of interfaces we’ve also created a large language model registry user interface where um apologies for the small text on

The right but you can enter clinical notes you can create prompts for the model that prompt engineering that’s a phrase you’ll hear a lot when it comes to large language models that’s really where we as Physicians come in to tune these models in the direction to answer

The task at hand and these interfaces now have a flexible architecture allowing us to evolve over time so the days of figuring out which 50 variables you want to put in your registry I think are over where now we can update these in real time moving forward and say you

Know what I want to look up these three additional things retrospectively on every patient that’s ever gone through the njr or whatever you uh are looking at from that standpoint so just in some early pilot data for this we uh looked at large language models versus the nurse

Abstractors that look at every surgical note here at Mayo Clinic to cure cre our own registry and in a couple of simple data points surgical approach for total hip arthoplasty the large language model is already outperforming our nurse abstractors it’s a little humbling and uh saddening I think to see that our own

Abstractors have a 4% error rate for something like surgical approach but it’s true when we look at um whether or not we used robotics or enabling technology in a case a large language model does extremely well on that at least in a test set of a th000 it was

100% accurate the nurse error rate is 2% so uh while the nurse rate is still good already these large language models are passing the Turing test for some simple variables as a proof of concept so now moving on to generative Imaging and I think this is the area

That I’m personally most excited about so what you’re going to see now is a video from our lab these are all 100% synthetic pelvis radioraps sort of morphing into one another just to show you kind of the broad spectrum of anatomy and pathology that can be generated on demand

And we’ve created a graphical user interface where we can ask uh the algorithm now to generate a series of radi graphs with parameters that we specify so I’ll ask the algorithm here to show me an AP pelvis without any prosthesis in place of a 65y old female

With low BMI and there’s a radaph produced in real time now I’ll have it change one variable to a BMI greater than 40 and you can immediately tell quite a very different softt tissue envelope on that patient we’re also working on a iety of other um conditional parameters for this type of

Tool such as severity of pathology of the joint uh presence or absence of an implant which I’ll show you here shortly and some other demographic parameters as well so to that question about generating uh radiographs with implants we now have an algorithm called th net this is published in Journal of arthop

Plastia for those interested that want to learn more but I think uh some of the implications for this it’s enabling uh Next Generation templating potentiating complimentary Technologies like augmented reality and Robotics and potentially even help us design new implants so let me show you how this works in another graphical user

Interface here you’ll see me pull up a real patients pre-operative AP pelvis x-ray I’ll then tell the algorithm I want to perform a right total hi arthoplasty I’m going to say I want to do that with an exit or stem and I’m going to let it choose uh the S tabular

Component of its choice and it’s going to make in a couple of seconds what it feels is the ideal post operative x-ray for that patient with implant now I’ll do an alternative example with a different patient using a uh cementless uh Emeral component um and you’ll see

That x-ray generated in just a couple of seconds there on the right so we put this algorithm to the test and what we asked our surgeons to do was to look at a real pre-operative X-ray and then two postoperative X-Rays one made by thet one that was a real

Mayo Clinic post-operative x-ray for the same patient they did not know which was which and they were asked on a scale of 0 to 10 to grade what we called the surgical execution basically how good did the surgeon do for that X-ray and quite humbling uh the surgeons in

Blinded fashion rated the synthetic x-rays higher than those of the male clinic surgeons so 9.0 versus 7.9 um a second metric here is we evaluated the percentage of acetabular components that ended up in the linic definition of the safe Zone the synthetic x-rays achieve that 97% of the

Time compared to 87% for the real X-Ray so at least on one very objective metric the these algorithms are picking up on what it uh at least perceives to be what good postoperative x-ray should look like some of the other areas we’re working on with generative Imaging are

In the domains of automatic view correction so this really allows us to understand Anatomy from various perspectives with a single X-ray and the perhaps the more exciting U implication of that is if you iterate that far enough down the road you can enable two-dimensional inputs to potentiate three-dimensional

Outputs So to that first question of view manipulation we have some algorithms we’re in the process of validating them but they’re already up and working so this is a real patient AP pelvis X-ray on the left and then on the right are four synthetic outputs for that patient and we’ve manipulated the

Position of that pelvis in space in the X Y and Z planes that’s what you see in the brackets those views my Juday views Etc um without any additional time in the X-ray Bay or any additional cost radiation Etc uh but I think what we are most

Excited about in the lab is taking this to the next step and here’s some of the early um prototypes here for taking two-dimensional inputs getting three-dimensional outputs on the top left are um uh drr versions of radioraps for real patients and then these are the three-dimensional outputs from that single input to the

Algorithm so nothing except that AP radaph and then we’re kind of spinning these in space to show you and uh I don’t have the real CT scans for these patients we’ve trained it on all patients that have had CT scans that’s how the model is learning to do this um

But when you match up the real patient CT next to this uh it is uh extremely high fidelity and we’re in the process of quantifying that so I can’t share that data with you yet but it should be coming soon so to wrap up here some challenges

And opportunities with all of these type of Technologies and uh I think it really comes down to leveraging the digital ecosystem that’s how we’re going to get the most out of AI and the challenge with that of course is integrating data it’s a extreme Challenge I would say

It’s a especially a challenge here in the United States where our Health Care system does a very poor job in contrast to the United Kingdom of talking to each other it’s so fr action but we need to integrate data across the spectrum of care pre-operative to inoperative to

Postoperative data to kind of create that virtuous cycle of feedback and then also importantly between institutions the opportunity there is for iterative improvement if you can create that infrastructure you can enable real time model updates and create a Federated learning architecture that makes AI really shine regulation and implementation is

Another big thing uh to think about this is a new frontier at least at the FDA it’s labeled as software as a medical device and these Frameworks uh are currently being worked out I’ve talked to some members at the FDA and they’re hoping to have sort of new software of

Medical device Frameworks out uh by end of this calendar year so that should be something to pay attention to uh the performance thresholds for clinical use are undefined that’s going to be a big question that we need to debate as the as the research and clinical community

In this space but the opportunity here is to augment what we do as Physicians and surgeons back to that concept of the digital co-pilot decreasing administrative burden making us smarter at our jobs and focus on the tasks that are really at the top of our training last one I want to focus on

Here is trust so big challenge of course so artificial intelligence algorithms get a reputation uh for being a bit of a black box they they can’t always be explained but there’s some tools that are now available to assist with that and there’s a lot of very smart people

Um working on that particularly in Silicon Valley so Physicians are uncertain um how much trust to put in these patients come in skeptical but the opportunity if we can get past that is to enhance care confidence so uh we want to emphasize models that come with tools like uncertainty quantification not just

An output from the model but an output from the model that says the model is 87% certain of this conclusion and here’s why for reasons X Y and Z and there are uh techniques that can do that and then also we need to to perform research on these tools and demonstrate

Superior outcomes we shouldn’t just take it at face value that because the algorithm told us that it must be better so a lot of research needs to be done to Shore that up so final take-home points uh AI will definitely change the landscape of Orthopedics over the next decade I think

That uh domains such as patient phenotyping algorithms large language models and generative Imaging will be catalytic Technologies in that effort and lastly I would say be prepared the pace of change is far too rapid uh to comprehend the chance that it will be disruptive is 100% And the chance that I

Can predict it is certainly 0% so with that I would like to thank you for the time attention and opportunity and would love to take your questions during discussion section thank you so much Cody I’ll hand over to Dr CA thank you Cody that’s fascinating uh

Talk uh without further Ado I’d like to introduce my dear friend Dr Grim who is a uh senior researcher in the field of movement and analytics and biomechanics uh he is also past president of eors European Orthopedic Research society and the current president of ior international combined Orthopedic

Research societies so go ahead Dr gri let’s hear your talk I think you’re muted uh Dr G do you want to unmute yourself thank you yeah s sorry the um so thank you Sur for the kind introduction and and thanks to sik for inviting me to this um

Beautiful lecture series and to be part of this distinguished faculty and giving me the opportunity to present something that I’m very passionate about um I hope you can see my screen it’s about the variables and how they um contribute to to uh digital mobility biomarkers in clinic and and

Research can can you do my slides uh progress yeah we can see I just hit the presentation mode so it goes into a presentation uh format yeah okay now I think now it worked um yeah this is this is this is this is not not a patient

Variable but it’s a symbol for me uh that a varable sensor can measure everything in in theory can be inobtrusive so that uh it does not affect what’s happening and the subject who’s using it or does not realize it and and I think that’s similar to what

Dr V AR presented with his sensors so this is the sensor and it’s inside a ball that was used in the football World Cup and when Ronaldo uh claimed that he uh took a head header goal it was later shown that the IMU sensor inside the wall showed that no head contact was

There so it was clear that this goal was not scored and I think this shows a principle of variable sensors in assessing patient or human motion and when I do human motion I mean this can be physical function so how well does a joint that has been treated perform how

Does it influence Mobility or disability in real life it can also be used uh to measure performance for instance athletic performance after sports injury treatment and we can look into physical activity Behavior because if treatment reinstalls a certain functional capacity it does not automatically mean that patients transfer it into real life

Behavior that we want to achieve or that they want to have so assessing human motion can be a diagnostic biomarker it can be a clinical outcome but we also know that uh physical activity can be a treatment so we have a treatment and an outcome that’s in the

In the same met in the same domain it can be used to predict uh predict disease risk for patient phenotyping and it’s highly personal I so I like to see it as a kind of an acto that goes along with other popular domains I think we

Must spend more time into this there is also a picture that adapted from my colleague benedi Brown shows that say a typical curve of a diagnostic metric that can be anything let’s say this is like maybe knee function or steps per day or so and you can see that normally

We would have some degradation of the normal performance which may warrant an early intervention when it becomes worse there is an injury risk then maybe there is a trauma or a treatment and there’s a rehabilitation and that can be different that we can have people who are have a

Fast recovery people who are not healing people who who are maybe overloading the construct and therefore cause reinjury but actually most of it is pled to us if we do conventional follow up with x-rays or questionnaire so this we can’t see this and this is all domains where

Variable sensors can come in so what what I think what is the beauty of varable sensors we can first we can uh do of course testing in the laboratory in the hospital uh doing big detail analysis so we can really look with many Technologies in a very detailed movement

But where varable sensors we can go out of the laboratory into the field or at home do long-term and frequent assessment versus normally only very momentary and episodic assessment and maybe we can also do for sure 247 behavoral component so not only what the patient can do but what he does do and

Not momentary but continuous and in that way we have big data that’s amenable for artificial intelligence exploitation that can be supplement the big big detail analysis in the lab and then we have maybe smaller star smart data in between that does something tells us about therapy achance for instance and

What other varable sensors that are available to us there is some some sensored family that I consider clinical or research grade devices that are developed for a specific T purpose and undergo scrutinous medical validation including in front of the medical agencies then we have these consumer grade devices that

Are that that are more and more available and then we have patients who bring their own Technologies to us so we will have a large range of devices which we can use for harvesting but if you do veral sensors normally you may have for instance planta pressure insults which

Can measure the plant pressure distribution the CR reaction force and of course in lower extremity healing after fracture for instance this can be very inform ative but the main uh sensor domain that’s used are inertial measurement units they contain accelerometers measuring accelerations and gyom meters that measure the angular rate of

Rotational um displacement and then there are sometimes fused with other sensor domains and when you use such a sensor and you use it in on various uh body parts and you have you fuse them into a model you can actually create a a full motion capture equivalent to the

Quality of a marker based gate laboratory that’s the current gold standard but you can see while this allows us to go into the field it’s not something that the patient may wear 247 but it’s it sets you free from the laboratory so normally you you may use

Only two three four five sensors maybe to focus on a single joint or to to to drive an EXA game application at home but in in fact most commonly are single sensor Solutions such as the ones worn on the wrist or the clinical CR sensors that are put on different body locations

And the body locations are chosen depending on the metric the the mobility metric that you’re most interested in and there are many sensors available but I think this discipline should not revolve around the the hardware but more the algorithm it should be device agnostic to drive the field forward and

Then of course we have the phone which is a sensor Powerhouse in itself and that can be exploited also and then we can also connect the phone and fuse the data I now show you some applications of variable sensor algorithm in um 247 patient monitoring to show you what kind

Of Novel Biometrics can be generated from this so this is a study where we used um a variable sensor for seven days for a week assessment in 100 patients after total hip replacement versus matched healthy controls and then you see that daily step count the mean value is different but it’s not significantly

Different there’s a huge variability because it’s an very individual U tra if you’re active or a couch potato but if you look in the distribution of these bouts you can see that total hip arthropathy patients have a different Behavior they have significant avoid the short boards and they significantly avoid the long boards

So these ones that you would would associate for instance with uh um Health supporting physical exercise avoidance of short walking and avoiding of long walking but because they are rare in both populations they don’t influence the daily steps but you can see that another condition like spin

Theosis has a very high impact on daily step counts using the same method this was also shown and another study independently performed where they did the same for NE osteoarthritis and total Nee arthroplasy postop again daily step counts no difference but again this typical avoidance of short BS was seen in

Thetis patients and the total knee replacement patients the sit St well there was a tendency for differences but it was not statistically significant this is another study that I like very much to to to show the power of using to assess patient mobility and outcomes this is comparing total knee

Replacement versus unic compartmental knee replacement from the database that was uh matched with gender Age and and BMI and in the patient self report like in many other studies there was no difference in outcome not in the course not in the Forgotten joint score and not

Even in the squash which is a self reported physical activity report and again step counts were not different but what was different and significantly above the minimal clinically important difference is the walking speed so the the step frequency and the usage of this of staircases so this is something that

Patients may not self report but a significant difference that this um treatment of the Lesser invasive treatment results in in a higher level of real life activity this another study that shows uh the added value of um real life by metrics assessed by variables where we compared uh patients with a low wear

After 10 years uh total hip replacement and patient with the high wear of the total hip replacement so the average wear and wear rate was three times higher in the high wear group so you would expect a much higher step count because you would think these are

Patients who are using their joints more but again daily step count was what was not different but what was different was the the walking speed and the um acceleration intensities and that was at first I would I could hardly believe this but uh in the followup there were

Three independent studies that did a similar study on metal on metal uh implants and they also found that that primarily uh intensity of the activities contributes to where not not normal walking this also where where where you can see uh that measuring time on feed

So the time out of bed in uh after total hip replacement in the in the hospital until discharge can be monitored and rapid recovery programs can be validated by this technique in the rapid recovery program the patients continue to wear the sensors at home and you could see

That once they were out of the hospital their activity actually drops or shows maybe the need for at home physiotherapy or activation this is a study where it’s not done by us but by by colleagues where a single sensor was used during post Rehabilitation after ACL where the

Where you can see that the knee flexion angle increases during the post of weeks and the asymmetry between left and right goes down as expected so this is just a single IMU and it connected to a phone for remote uh monitoring so we can also have longitudinal granular detail of of

Rehabilitation this another study conducted by my colleague bentic PR who’s present they they use the planter pressure insole to monitor compliance to uh post fracture loadbearing protocols and what they showed was that there was a very compliance to this showing that patients cannot uh Implement these postoperative loading protocols but it

Was also shown that did not affect the out so maybe this can be interpreted as from consumer variables that the patients brought themselves the beauty of this approach is that you can collect the the metrics also from the pre uh traumatic phase when you harvest the data so this is 57 highly unhomogeneous

Fracture patients but it was possible in many cases to identify the time of fracture and a rehabilitation phase uh back to normal but there were also people who did not show this and where maybe non-healing could be uh uh derived or or bad healing could be derived from

This it can also be used in the upper extremity and can use a phone for instance to replace the gometer but you can also monitor in real life activity differences between healthy controls and uh shoulder patients where you can see that the asymmetry between a dominant and a non-dominant side is affected

When um patients are injured or undergo treatment but while while I have while I’m a big fan of variable sensors the recent advances in computer vision have en AED us to to use a single camera Consumer camera uh with computer open source computer vision models to generate for certain gate sorry gate

Metrics a similar quality of motion capture as in in a 3d motion cture in the laboratory so maybe some sensors may even be uh replaceable by uh this uh approach so coming back to the learnings uh of of this session the variables are available to assess many dimensions of

Physical function activity they can use be used for Diagnostics outcome assessment remote monitoring and patient self management previous year inaccessible insights are possible into early disease detection progression or Rehabilitation there’s much Choice little consensus and standards and validation is lacking so we have to choose between requirements and

Usability and in the future we will we should go for devis agnostic me methods bring your own device harvesting and computer vision thanks for your attention thank you thank you uh Dr Grim so um we have a little time for a couple of questions I think one of the

Questions that came up for Dr Braun was uh how does The Strain change as the fracture heals and how can we intervene as surgence so so the idea is the the the implant measure The Strain on the plate meaning that the strain over the time of fracture healing that is measured by the

Plate or by the sensor on the plate should decrease because the load is shared more and more through the through the fracture and the healing bone so the load on the sensor should decrease when you have a fractur healing and then there’s been some animal experiences uh

Experiments run and you can clearly see the animals that do not not show this drop in the relative implant that they are actually going into a non-union and the sensors picking it up and the idea is that you could a have an an earlier intervention um and and really detect

Patients that are at risk for a delay in in fractur healing earlier than you could with uh with traditional measures thank you thank you um I I got a question from the uh the crow I’m I’m going to combine a couple of question question that’s okay Cody regarding your

Your fascinating fantastic talk it’s just it feels like the future’s already here it’s just not evenly distributed yet so the question comes regarding the risk prediction calculators you mentioned um the question is how have you integrated these calculators into your practice uh what challenges have you had any in that process and then

Someone’s noticed that you put a radio graph into that calculator as well is that being analyzed to modify the risk score or advise as surgeon what implant to use or approach to consider so there’s a bunch of things really just clarify a bit more about those calculators very good yeah thanks for

The question CTIC and those that came through the chat so um specifically the dislocation and paraprosthetic Fracture risk calculators are now live within Mayo Clinic so Mayo Clinic has campuses all over the United States so um for variety of reasons it can’t yet leave those walls and it’s also giving us a

Chance to test it uh in real life make sure that it’s working properly that uh the data that we’re getting is is uh is solid and getting feedback from surgeons uh on whether or not it’s helping them whether or not they’re changing their management as a result of that uh all

Questions that are important so the calculator um does take into account all those demographics and comorbidities it does take into account a pre-operative xray and the X-ray will be assigned a uh percent percentage importance in the decision by the algorithm so um for some patients you know that ends up being a

Very dominant Factor but it will tell you that um so that you know how much it’s uh how much it’s giving weight to it um that algorithm is separate from what I showed for Implant selection that th net that’s a different algorithm Al together that will simply with no other

Information take a pre-operative x-ray and then you can choose to let the model select the implants or you can choose the implant and it will show you what it believes the quote ideal postoperative xray should look like um so that’s how it kind of delineate those brilliant

Thanks Dr ws and we’ll stick with you for one more question before moving on to a couple for Dr vanak as well it was regarding the generative Imaging the 2D to 3D Transformations you mentioned some real world applications there could you get a little more into the Weeds about

How that’s useful for templating for Robotics and uh for AI the three things you mentioned yeah absolutely so uh you know I think with I’ll differentiate it into two with view correction and true 2D to 3D generation so view correction I think uh the two biggest applications at least in my mind

Are uh getting multiple views of the anatomy uh from a single x-ray so if we get a five view series of the pelvis for example we can do that simply from the AP pelvis so now we can get AP pelvis Juday and Inlet outlet that type of

Thing um but I think the more powerful application for that is in theater with fluoroscopy so if you’re doing a trauma case in particular you can automatically Envision the value of that for not having the radiography Tech you know uh that’s having a bad day and maybe you’re

Having a bad day and now we’re compounding problems and so if we can um have the CR arm put in one single position and get multiple views um you know to check our reduction to check our starting point that is going to be very powerful now 2D to 3D I see that is

Having bigger implications for preoperative planning on complex cases and interfacing with robotic so for robotic knees for example one company famously gets CT scam before most are trying to move away from that but you lose something when you don’t get that um there’s a lot of push back at least

In the United States to get CT for primary knes so I think this is a way to bridge that Gap get the best possible information for the patient to enable robotics without all the downs and there are many of the CT cost radiation access Etc thanks Dr W certainly I’m Keen to

Get the alignment and some morphology stuff from the technology you’ve presented and trying to get rid of that pointer if that takes 10 minutes to you mark everything down so I think that the future is promising thank you very much um Dr vanar if wouldn’t mind uh asking

You know actually a group question to you um Dr grim and Dr Brown so the flavor of the questions are you know about moving on from traditional metrics of measuring outcomes because you know x-rays tell half the story non-union may not be caught until too late or much

Later and as as you mentioned uh Dr Grim that modern proms have sealing effects so I’ll ask each of you this question in in series Dr Van arle can answer first in relation to the devices that you talked about what is it right now that’s preventing these Technologies from being

Mainstream is it cost is it regulations or is it something else Dr vanle first that’s okay um I guess there’s no actual barrier it just the work needs to be done so the stuff is invented and then you need to translate uh any any implant where you want to influence clinical

Decision- making would be a class three device and that comes with a higher regulatory burden to take it through and so there’s a cost so there’s an investment raise and then there is a time to do the trials but that that is the barrier and that is the barrier

Facing all the devices and particularly if you want to chase that clinical decision- making Factor you have to be class three and that will slow things down in a clinical trial so there will be a safety trial like we see with the aactor Monitor and then there’s a

Pivotal trial where you’ll start to really be able to do those marketable claims and make it a product great thank you and and Dr Brown you’ve been involved with development of this device it’s clearly an AO device that’s being tested right now what are the next hurdles before it becomes you know

Something I can get on my shelf like Dr faka just mapped out this is the process that this will have to go through um I was quite surprised leading up to us being able to implant it the the process was quite smooth so uh I

Initially I and I was the PCI you know I was in all the documents and luckily AO has a lot of people that are really good at writing these things so really I I was mostly I was just okaying them but but but still I was um

Surprised at how well it was received by the by the regulatory um authorities in Germany um that did raise questions but they were answerable and they were um showing quite a high understanding of of what they’re trying to achieve with it so um uh I would have imagined the

Process to be worse so so there’s an obstacle I agree but it’s it’s a manageable one great and Dr Grim I think you know it was a nice symbiosis for what you said regarding partial knees you showed that problems show no difference whatsoever when you start measuring things that matter to patients

In terms of their actual function it’s it’s a step change so how do you propose that we get Senses Into more patients so we actually see the differences and then convince surgeons to do more of the the stuff that’s conservative or just a different track Al together yeah it’s

It’s really like as as you question said these technologies have been around for a long time but they have not yet become common place and and I think one of the hurdles is um also the the plethora of devices that are around the lack of standardization the lack of consensus uh

And and and that um and in the future I think it will be that more and more people already have wearable devices and may not want to add another one so this patient compliance and usability is is a challenge so maybe one of the promising Futures to make this more common place

Is to actually uh use and harvest data from devices that the patients bring themselves and maybe improve the the algorithms that run on it I mean now they are let’s say more or less glorified step counters but with with the use of uh AI we we we we can also

Detect maybe more specific and meaningful outcome so for instance it’s now possible to also detect um to a certain degree the surface on which patients are working so if they are taking the steps or walking on on ramps and so on and and that that can be an

Interesting metric that you could derive from from something so I think it’s probably the the main is is is usability and maybe to to overcome this is to harvest What patients already or subjects already bring yeah great thank you so much and I think we have probably time for one

Quick question that’s come through for for Richard um it’s uh it’s about the accelerated development and the reduced cost and the animal burden that you’ve shown with the use of a bioreactor as a first step um is that something that’s been developed in hous Imperial or something commercially available to test

Implants as a next kind of phase to accelerate development before clinical trials uh something we really saw coming out DeVos um they developed ways to keep the bone viable and then we saw the opportunity where we have really high development costs with animal trials particular if you want to influence bone

Turnover or you want to look at things like how external factors like bacteria interact with a living tissue like bone and so we just we borrowed the idea built it up and it led to this bioactive approach and we found it’s just we do it because it’s cheaper we do it because

It’s faster and we do it because it’s really accelerated our internal devel velopment and then there is still value in that full animal trial you learn a lot by going all the way but certainly for those early developments it’s really accelerated the tech development and so that investment on this bioactive method

Really worth it and we’ve published the method we’re not trying to hide that because we see the value in that independently of the tech where we’ll be more C brilliant thank you so much Dr Van so that brings the meeting to a close time is up uh I’m so grateful we

Are also so grateful listeners included for spending your your time what’s going to happen next is we’re going to flash up this QR code I’d be grateful for all the listeners and viewers to to flash that on your phone camera and then provide us with some feedback and your thoughts and get a

Certificate of course for for CPD and your points uh and and and so now I close the meeting thank you Dr Wes Dr vanle Dr grim and Dr Brown and of course my co-chair Dr C anthon okay have a have a great day thank you thank you everyone byebye