Well, first of all, thank you for everybody being here tonight. Welcome to this webinar. My name is Sylva Janska.
I'll be your chair for tonight. And, you know, obviously, thank you to, to all of you being here. I thank you for, for the sponsorship from Single Pet for making this even possible.
And thank you to Neil Shaw for joining us and and speaking to us and I'll just like to say a few words about who you are as an intro, so I know you've been always somehow intertwined with veterinary medicine. You were born in South Africa, but then spent a lot of your time with your, fellow two brothers at your father's small animal clinic. In in Florida where you actually went to vet school, you went to the Florida College of Veterinary Medicine, and I believe you also stayed there and did did your internship or residency there, and you're now a diplomat in the American College Veterinary Internal Medicine.
And you founded the nation's largest group of specialty hospitals, Blue Pearl Veterinary Partners, and after that served as the first chief medical officer from Mars, Mars Veterinary Health Group, which is, wow, just, you know, already wow, and, and then subsequently in 2018 you went off and founded single pet. Which is an independent company dedicated to bringing the benefit of advanced artificial intelligence assisted diagnostics to veterinary professionals, anywhere, any location, and so, as we can probably all all guess, the technology enables obviously advancement of care, advancement. Of, you know, standard of care, setting new standards and obviously the immediacy of information to the practitioners as well, which is just amazing and, and I'm very, very glad that you found the time to speak to us today, and I believe the title of your talk is Veterinary Ray, the new Challenges and New Standards.
And before I hand over to you, you'll have the floor in a second. I just want to say that, this webinar is being recorded, it will be available tomorrow from 5:30 p.m.
UK time. It is also a, both C and CPD accredited, and the free certificates will be available also from tomorrow and if you have any questions, guys, then please post them into the Q&A box. They might get lost in in the chat, so just go to the Q&A box and post them there, and we'll take all of your questions or as many as we can within the time, at the end of the session.
And so without further ado, over to you, Neil. Sylvia, thanks. I appreciate, appreciate the introduction.
Appreciate everybody making time. I think the majority will be in the evening, some in the afternoon, and, and a couple in the early morning. So thank you.
Thank you for joining us. I, hello from Dallas. It's, it's warm out.
It's, in the vicinity of 100 °F, somewhere around, I think, if I'm correct, 39 °C, 39 at times approaching 40. So, hello from warm Dallas. If you guys are in cooler climates, I'm happy to come and visit.
. So I'm going to speak for about 40 or 50 minutes. We will certainly make time for questions afterwards, and talk on radiology and specifically artificial intelligence use in radiology. It's a process that I've been involved in for the last 4 to 5 years now.
It's been an absolutely fascinating process. Pardon, pardon my bias. I, Enjoy it more and more, with a, as Sylvia said, growing up and effectively growing up in a small animal practise, being heavily involved, in my previous career in specialty medicine.
I see AI as our future. I, I genuinely do. It's, we read about AI, I read about AI, didn't know what it really meant, kind of heard about it in cars and different fields.
And only when it was applied to something that I knew so well did I see the magic, and it, it indeed is magic and is of significant benefit to us as a profession, us as clinicians, to healthcare teams, to pet owners, and to no doubt, and most importantly to the pets. So, I'm going to jump in and start talking again, by all means, happy to, happy to answer questions afterwards. So what do we do?
What I'm going to reveal is what we did prior to, being available is I got together with some tremendous engineers, really, top, just exceptionally good, not to our, our engineers is putting it lightly. Well, one guy in particular who is an architect. He's an architect, his mind works in ways that I'm still trying to understand.
And I posed a question to him. He's a dear friend. He actually lives in Australia.
Ben, Ben, can you, design a system to interpret radiographs as well as we can read with human eyes? And he said no problem, and I challenged him. And he did it.
He absolutely did it. But now there's a whole team of engineers, there's lots of veterinarians, there's lots of specialists involved. But the core essence is we're able to take anything that is repetable, constant and repetable, and turn that into software.
So, in this talk, I'll discuss what initially, our approach, what didn't work. Then I'll discuss what did work. I'll talk about the current capacity today of AI in veterinary medicine.
I'll go through 10 case examples to give a clinical perspective on how this works in day to day seeing cases. I'll talk a little bit about the economics of quality medicine, of good quality medicine, and mention a little bit about what I anticipate the future to be. One final thing is, somebody I work very closely with, head of education for Signal Pet, Bass.
Bass is not in the same room as me, but she, she is forwarding the slides. So occasionally you'll hear me say, Bass, please forward the slide, and then we'll go to the next slide. So Bass, if you don't mind, we'll go to the next slide.
Thanks. So what didn't work? The initial thought, and this is now going back several years ago, when folks design artificial intelligence systems, the word that comes up quickly is data.
It's all about the data. And the misconception is you simply feed data in. To the computer, the computer generates an algorithm, and then that algorithm can be used in, in, in place and for more.
The challenge is we couldn't get it to work, and what was it? We had lots and lots of films, lots and lots and lots of films. We had the radiologist reports from those films.
And we fed those into the system. And consistently, we could not get the algorithms to lock and could not understand why. It simply didn't work and it was an initial, it was an initial period of a, of a frustration.
Went ahead and said, gosh, let, let's reset reset here and pull a number of specialists. And have them all look at the same films. So we're not looking at different films and with different reports, we're looking at different reports, but all in the same films.
And again, we could not get the algorithm to lock with the additional data. So what, what happened? Pa, next slide, please.
The short of it is that they There's the old expression, garbage in garbage out. And we used to discuss that, with pathology samples. It also, it's also true with radiology and creation of algorithms.
It's not that necessarily that each radiologist's report was wrong, but there was a definitive lack of consistency to the reports. And because of that lack of consistency, the, the computer could not identify patterns in reporting. So what we did is we took the hard process of going through, and on thousands and thousands and thousands of films, we reinterpreted those films in a very structured way.
The structured way of interpreting films was not necessarily available. We had to develop our own protocols, which we did. And as a result, we created millions and millions of data points.
It was a long process. It was a tedious process, and it frankly was an expensive process, but it worked and it worked very well. And I remember where I was at the moment.
And went through a thought process that initially our goal was to create an algorithm that meets the, the quality control of human eyes. And what happened reversed is now our human eyes were working hard to meet a consistency as determined by the computer systems. And it's, it's a humbling experience.
It's a humbling thought, but it's an amazing thought as well, is that, that we now had a real quality control on what we were actually doing. So, because of the challenges to revisit, it was because of challenges of variability and interpretation. That was our most significant issue to overcome.
We could do that by creating consistency to the data. And interestingly, when we did that, we realised that the computer is very, very, very good at interpreting films. It doesn't look at it with human eyes.
Radio radiographs are pictures. Pictures are pixels, pixels are numbers. The systems today are excellent at analysing patterns and numbers.
To the point I even films that an average practitioner can take a look at and say, yeah, I'm pretty sure I can see something there, which may not be a picture-perfect film. It may be a little bit rotated, it may be a little dark, it may be a little light. It may not be picture perfect, but if the, if the litmus test, if most clinicians can feel that it, it's, it's pretty much diagnostic.
That comfortably is diagnostic for an AI system. And so variability of image quality actually wasn't as much of an issue. Variability and interpretation was the key issue to overcome.
May I ask the next slide, please? So what did we do? We created the, the largest known database of veterinary annotated images.
We actually patented the process. We patented the process of, of how to transfer information from plain film radiographs to create the creation of algorithms for that. And that, that process is state of the art for human, human medicine as well.
We have an accuracy that is confirmed by lots and lots, millions and millions of data points, revisiting, revisiting, revisiting, not only in the past, but at present and definitively in the future. So we constantly reassess, constantly have a quality control programsuring that what we see is nothing but excellent. At the same time, as it grows, the system improves, right?
And it gets better and better and better. It's absolutely amazing. And this and this quality control process that I'm aware of that doesn't have otherwise exist.
It, it is a new, it is, it is a new standard. That's next slide please. What we first did Was identify what are the most common syndromes seen on a radiograph in either primary or emergency practise.
And we split that up into 3 panels, the thorax, the abdomen, and the skeletal system. And so then what you'll look at here are the most common findings. So we looked at vertebral heart score as a measure and the computer calculates it as a measure of cardiomegaly.
And then if you are concerned that the heart is decompensating, you, you would look to and it's left-sided disease, of course, you'd look for left atrial enlargement and comfortably, the computer can identify left atrial enlargement. Then there's, especially on emergency, there's the unusual ones such as esophageal distensions and esophageal foreign body, biphromatic hernias, pneumothorax, pleural effusion. A cranio ventral pulmonary parenchymal pattern, suggestive of aspiration in most cases, a qua a dorsal parenchymal pattern, suggestive of heart failure, masses.
I'm moving, going through a little bit quicker. 22 populations of small intestine, consistent with mechanical obstruction of the small intestine, suggestive of surgery. GDVs, gastric distension.
Big livers, prominent spleen, especially in cats, which is abnormal, limited abdominal detail, either in a very young animal, very thin animal, or, if there's a bleed or such, fluid in the abdomen, on the urogenital panel, mineralization in the kidney, calculi, anywhere in the urinary system, big kidneys, small kidneys, prostate being visible. And in the US more male dogs are neutered. I believe in Europe, it's, it's more common, to, to not be neutered.
And then, on the skeletal system, the common findings on the skeletal system, narrow disc spaces for intervertebral disc space narrowing, spondylosis as a result of, chronic instability of the vertebrae, and then on the forelimbs and the hind limbs, looking for litic lesions, looking for fractures, looking for specific things such as effusion in the stifle, suggestive of a cruciate ligament was such, and so on and so forth. And it's worked very well, and I'll show you. And then finally, there is a dental panel as well since dentistry is, is, is done with an increasing, frequency, specifically dental films, and we look for the most common findings within dental films, which are forcation bone loss, periodical lulucency such as a periodical abscess, resorptive lecision.
We added retained roots. We started to see many retained roots, fractures, attachment loss, and And then identified such as canines and PM4s, key, key teeth that we want to retain, that have pathology that either ideally, we don't treat with a, a, removal as the first line of therapy. So, Bass, could we go to the next, next slide, please?
So now we're going to go through a number of cases to show you how to use AI when you're interpreting radiographs. And the key here is remember the AI, I say remember, but the AI is almost instantly available. So a radiograph is taken and then normals normals and abnormals appear on the right-hand side.
So what will initially happen is we'll show the radiograph and then Bass will go to the next slide and we'll show what happens when you look at the radiograph together with the AI. And so we'll get just 1 or 2 seconds to look at the radiograph. Everybody can hopefully, form your opinion on the, on the film, and then we'll, we'll show with the, with, with the AI information next to it.
So this is, this is Bella. By the way, names have been changed and no, I'm pretty sure no owners' names or hospital names have been listed. They're all real cases, but no, they, all the films have been anonymized, as, as is appropriate.
So what Bella's abnormality, look for a second or two, and then look at the AI. So you look now and then Bass, do you mind giving me the next slide? And then when you look with AI, the AII certainly circles the heart because the, the heart is abnormal and identifies one, there's an increase in vertebral heart score.
So without a doubt, it doesn't quite show it on this slide, at least on my, on my screen. It calculates the vertebral heart score and that is, that is elevated. It's not a 100 VHSs aren't 100% perfect as far as assessing cardiogaly, but they're very good at putting you in the ballpark.
So the VHS is elevated. There, there is a cardiogaly which is true in this case. And if I look on the right-hand side, best, if you, I don't, if your pointer works, could you show the four red lines by VHS?
Thanks. Thanks. So what we see here are 44 red lines by VHS.
And again, this all occurs instantly. And what it says is that with 4 red lines that the computer has absolute confidence that there, that this is abnormal. If there were 3 red lines, then the computer would have abso would have most likely say, most likely that this is abnormal.
Conversely, if there were 4 green lines, the computer is saying. With absolute confidence that this is normal. And if there are two green lines, the computer is saying, most likely this is normal.
So what do we see here when we look at a, an AI interpretation of this film, which ultimately clinicians always have to review, don't, don't let the AI report stand by itself, always the clinician reviews. But what we see here, of course, are an increased vertebral heart score, cardiomegaly, yes, and left atrial enlargement consistent with left, left-sided decompensation. And, absolutely, that is correct.
We have left atrial enlargement. Unfortunately, I can't circle it, because I can't control the mouse, but in that, that dorsal hylar region, you can see that exact, thank you, Bass. Thank you, that the left atrium is enlarged, and, pushing up on the, on the trachea.
Importantly, it goes, and then what the computer will do is a scan for everything every time. So there's no evidence starting at the top. There's no evidence of an esophageal foreign body.
There's no evidence of pleural effusion. There's no evidence of a pneumothorax or pleural gas. There's no evidence of a thoracic mass, and so on and so forth.
All right, next, next slide, please, brass. All right. So hopefully it's it comes through on my screen.
Hopefully it's coming through on your screen. This of course is a kitty cat. We can take a quick look, form your impression, and then look at it with AI.
So what, what does this cat have? It's a kitty cat, of course, it's a, a lateral radiograph and bass, next slide, please. And what do we have?
We have 4 bars of a bronchopulmonary pattern, which is consistent with inflammation around the airways, which is consistent with feline asthma, most likely. So very quickly, it'll pick up feline asthma and separate that from other parental diseases of the, of the lung. Next slide please.
All right. What's the abnormality here? I'm gonna take a sip of water and we take a look.
Everybody can look for a second and then Bass, could you go on? So, we have 44 red bars of a quarto dorsal perenchimal pattern. When, when we were looking at perchial lung patterns.
We had to figure out ways to classify them. And there are many, if you, if you read some of the radiology textbooks and listen to some of the lectures, there are, of course, are many, many different ways to classify lung patterns. And, and each has their individual preferences.
We had to find a consistent protocol that would work across the board. And the consistent protocol that worked across the board was the geography or specifically Which lung lobes were affected. So in this case, we see the quarter dorsal pachimal pattern is consistent with infiltration of fluid in that in the quarto dorsal lung lobes, which is consistent with congestive heart failure.
There are, of course, there are other things that that can occur as is listed in the description. Most commonly, it's a CHF. Sometimes young dogs will chew an electric cord.
And, they can also get a a CHF looking pattern, but, most likely, and when it's in combination with, carddiomegaly, in which this, this patient does have cardiogaly, it's, it, it's consistent with, with congestive heart failure. Next slide please, perhaps. So let's take a look at this slide.
This is, this is Bentley. I don't think the name is Bentley again, it's been anonymized, but we'll go with it. So, Bentley, Bent Bentley is probably coughing.
OK, that. And again, these are all real cases. So before we discuss the quarter dorsal lung field, and this, I don't know if it's coming out well, but this is infiltration of the cranial, cranial ventral lung fields of which the middle lung lobe.
Is included in the cranial ventral lung fields. And infiltration, and actually, you can actually see a bronchograms in this case, infiltration of the cranial ventral lung fields can occur in many diseases. Most commonly when it's isolated, it occurs due to aspiration, typically aspiration pneumonia.
And, and this, of course, is a case of aspiration pneumonia. But, I don't know if you feel comfortable identifying the air bronchogram. Guys, Bassy isn't a veterinarian, but, but she's been hanging out lower, lower, lower.
Lower. Yeah, right. Yes, right there, they have bronchogram.
So, again, the computer will pick it up very easily, and infiltration of the cranno ventral lobes, most likely with the aspiration. Of course, you can have things different. You can have collapsed lung lobes, which will cause consolidation.
You can have neoplasia in there, of course, but we're looking at common things occurring commonly. The 80/20 rule, that 20% of the cases will occur 80% of the time. And we, and we're gonna go after the most, the most common ideologies.
Next, next case mass. So this is a case I give everybody, it's looks like it's also . A kitty cat.
And, what, what's wrong with this, this kitty next, and then we'll look at it with AI. And what we see is at the, what's been isolated here is this pleural gas or no doubt a pneumothorax. The heart is lifted up off the sternum, or the lung, or lung fields have retracted from, from the edges, and this is a pneumothorax, presumably a hit by a car, it could be something else, but, it is a, Are, are you guys having difficulty hearing me?
OK, I'll keep I'll keep going. I was just to say there is a couple of people who who said that they have issues, but majority are fine so the issue is probably with those couple of people that raised it as an issue. So guys, if you can either redial sort of in so sign out, sign in or check your internet connections please, because majority of us and and thank you all all you who is working for for for saying it's working.
I, I think it's, it's your end, if it's not working. Thanks and I'll I'll I'll make a point of stepping closer to the microphone here as well. So just just in case.
So this is, this is a, this is a patient with a pneumothorax, and the, again, the lung lobes have been contracted, the heart is lifted off the sternum. There's air, air in there. And, and the beautiful, the beautiful thing about this is if a technician is taking radiographs, this is a patient who's presumably just having trouble breathing.
The, the tech veterinary technician doesn't have to wait for the veterinarian, not to do any procedures, but the veterinarian, veterinary technician can take the radiograph, receive the report, identify, the report identifies as a pneumothorax, and can start setting up for a thyraocentesis to remove the air from the pleural space. And then confirm, not do the procedure, but confirm with the veterinarian. If the veterinarian's busy, say, doctor, when you get a chance to look at the films, confirm that it is indeed a pneumothorax.
And by the way, I have the thoraocentesis, equipment and a needle set up to, to pull the air. And so to aspirate the air. And so it becomes a much more efficient process when, when the healthcare team is involved and everything doesn't depend on just the veterinarian alone, of course, the veterinarian is ultimately the only one who makes the final, the final assessment.
Next case, please. So I'll give it a, again, I'll give it a few seconds. And Everybody see the abnormality here.
And clearly in Bass, I don't know if you see it. So what we have is a distension of the oesophagus. We, we no doubt see air within, within the oesophagus.
And at the same time, and Bass is circling it, we see esophageal foreign body. So we have an esophageal foreign body, more common on emergency than in primary care, but, but certainly, but certainly there and present, and again, the AI will pick it up and confirm it very quickly. Next case, please, Bass.
So we're moving on. This, this dog is, Winston. So Winston has abnormality, actually a couple of abnormalities.
We'll focus on one, in the abdomen. Next slide please pass. So Winston has, of course, a hepatomegaly.
So we see, we see the liver extending, thank you bass, extending well beyond, the, well beyond the ribs, the cartal arts there, displacing the stomach and, and others. And it will consistently, AI can consistently pick up hepatomegaly, just as we've, we've all trained our own eyes, our own eyes to do so. Multiple causes of hepatomegaly, oops, can we go back, please?
Thanks. Multiple courses of hepatomegaly, of course, if this were a cat, hepatic idosis, lymphoma, if this were a dog, certainly Cushing's disease and, and, and other syndromes, never mind abscesses or neoplasia, so many different, many differentials, but the case can be then taken and worked up, worked up from there. Next case, please.
OK. Do you guys see the abnormality? Oops, OK.
So, so interestingly, the AI picked up two abnormalities here. So certainly the most obvious one is if we look at the bottom, is UBC or the presence of urinary bladder calculi. So that, that is certainly there.
But also, if you look just cranial to the bladder. Just in front of the bladder bats, there's a, there's a lot, there's a lack of detail. So it's this kitty cat.
There's a lack of, there's a lack of detail right there. And presumably, I don't know if it was due to a cystocentesis or there's a little leakage, there may have been a little leakage of urine from the bladder, because this is otherwise a cat that has a good bit of fat, and we should see an increased detail. Again, the AI systems can differentiate between good, good serosal detail and poor serosal detail.
And so, this cat, this cat has two syndromes. It has the urinary bladder calculi, as well as a lack of serosal detail right around, right around the front of the bladder. And the AI will pick it up.
Next, next case. So what is the abnormality here? On rocker.
OK, that. As well. So this is another case, probably, probably due to a neoplasia, but there is a significant loss of abdominal detail in the cranial abdomen, due to a few accumulation of fluid.
There also some other abnormalities. We'll focus just for the sake of this conversation on the lack of detail in, in, in the abdomen. Next case pass?
And then another one that is, that is common. So this is, this is, of course, a stifle joint. And what do, what do we have on the stifle?
Often there's a question when a, when a dog comes in acutely lame, is it a ruptured cruciate ligament or partially ruptured cruciate ligament? And what do we look for? Of course, we can't see the ligament.
We look for the evidence of fluid, and often the evidence of fluid is displacement of that interpatellar fat pad. And the AI is excellent at identifying positive versus negative fluid within the stifle joint, which will be supportive evidence of the rupture of that, of a cruciate ligament. Of course, other syndromes as well, but most commonly, especially in one joint, that's what, that's what it would indicate.
Next slide, please. All right. So there's, there's 10 examples of AI use in clinical practise.
It's very straightforward. The advantages are that one takes the films and then the clinician pretty much on a, immediate basis. It takes usually within 10 minutes, most, most commonly less than that, is able to assess the radiographs together with the AI.
And it's been shown because we now have it out in practise to be of significant aid to the clinicians. And then in very progressive animal hospitals, the technicians will actually go ahead and read the report when, when the clinicians are busy, review the report. And then bring the results of the report to the clinician to give a final assessment.
It further increases the efficiency of the workflow from, from, from waiting for the clinician to get around to looking at the report, looking at the films. The next, sorry, can you go back to that back one side please. So then, then, the next, the next approach would be to look at what this actually does within a veterinary practise.
And when designing the system, the theory was, just as the slide says, good medicine equals good business. So business and medicine are tightly intertwined. One can't practise medicine, medicine, medicine and be independent of cost because it won't work.
At the same time, one can't run a veterinary hospital, for the bottom line, bottom line, bottom line from a business aspect, ignore the medicine and expect that to work as well. So, in just about all veterinary practises, there is a tight relationship between the quality of medicine and the quality of business, and ideally, both respect each other and in order to have a healthy practise. So in this case, the theory was that if we improve our medicine, so how do we improve our medicine by providing a greater consistency of care with radiographic interpretation by offering it at a low price.
The price point is low, actually very low, and increasing the value to the pet owner so they can get results quickly and confidently, then it will result in better practise and It actually proved itself out. I'll go through some numbers, but it proved itself out very early and very early on. The next slide please, Bass.
So, excuse me. We're now in, we're now in lots and lots and lots of, of, of hospitals. And even though I haven't found set standards, what we're seeing is for a, a high quality, we, we look at two groups, primary practise and emergency practise.
So in general, I'm going to say this in general, this is not published, but it's looking at me in thousands of hospitals, I see this trend very comfortably. For high quality primary care. This is, and this is actually is mostly in the US for high-quality primary care.
Generally speaking, 5% of visits will receive the benefit of radiographs. So 5% of patient visits, 5 out of every 100, will receive the benefit of radiographs, and it's, you can simply calculate the number of in, in any given month. You can simply calculate the number of radiograph studies divided by the patient visits.
And, most are lower. In folks who've embraced radiology, it will be about 5%. In emergency practise, The, the, the folks that utilise radiology well are actually taking films on 20% of patients that come in.
Specifically The number of cases that are presented in the denominator on the bottom, and the number of radiographic studies on the top, typically it's 20% on, on strong emergency practises. Again, these numbers aren't typically looked at. I haven't seen references to it.
The, what, what we now term the radiographic utilisation rate, but there is consistency in the patterns, very, very much consistency in the patterns. The interestingly on emergency, most folks in emergency don't feel that their rate is actually higher, that they take radiographs on, gosh, half, half the patients. It's not uncommon to say half or three-quarter of the patients that come in, and it's actually 20%.
And once they calculate it, it's much lower than, than they appreciate, they being the, the, the, the primary folks involved with the practise, the clinicians, and, and such. Next slide please, Mass. So we did a study.
We did a study on 23 practises, 23, 24 practises. It was We looked, we introduced Signal Pet to, to these practises. They, they, they weren't specifically chosen except a willingness to participate.
So there was no, no specific demands, generally known as good quality and a willingness, a willingness to participate in the study. We looked at At their numbers 2 months prior to adding signal pet. And then 2 months after signal pit.
So over a 4 month period, no, no, no push, no, just simply introducing signal pit into, into the practise with very little follow-up. Just make simply making it available. To read, to read the films and give clinicians confidence, so clinicians could, could give owners, pet owners more confidence and therefore the owners compliance or following the recommendations with the, for the clinicians to take films was higher, simply said.
And we looked, and it turned out to be a total of about 50,000 visits, 28,000. For the two months prior, and 27,000 for two months after. And, it's, and we normalised, we normalise the numbers because there's a little bit of variation, 28,000 versus 27,000 visits.
We looked at the utilisation rate and then we looked at films actually they sent out to a radiologist. Next slide, please, just to summarise. So what we saw was that over the 1st 2 months, and actually, we've looked at it subsequently, and the numbers are, are increasing beyond that, but just in the 1st 2 months, the extra utilisation rate increased by 10%.
So 10% more films were taken the first two months of introducing AI than prior. The number of films sent for radiologists consultation actually decreased, which, which is an interesting number. What we find is folks who heavily rely on radiologists consultations usually send out more selectively for radiologist consultations, and folks who've never used a radiologist actually will send out a little bit more.
On average, around, around 10, 15% that seemed to go out for radiologist consultations. And the customer charge on the study was increased by 7%, just, and that's the charge for the AI. So 7% on just the, on the film charge, and the, the net effect was a 14% increase in the radiology profit or EEA.
That's in primary care practise, on an emergency practise. We looked at 2800 patient visits. And we actually, in one, remember I said that about 20%, there's a 20% utilisation rate of radiographs in emergency, which is lower than most folks think, but this clinic was actually utilising 30%.
So they initially were utilising 30% and they said there's no way you're going to increase the number of radiographs taken because we take on everything, we take a lot of radiographs. Simply by introducing AI, giving the clinicians confidence, give it, which gives the client's confidence, which increases compliance, which increases medicine. What we had was actually a 12% increase in, the, the ratio of radiographs per patient visits, the radiograph utilisation.
We increased based on the revenue, net revenue of 16%, and we increased the total radiology revenue as a, as a, as a factor of the total visit revenue on emergency by, by 24%. The, the third case study, I'm not going to present numbers. These numbers are so consistent.
They've actually, we've, we've seen repeat of these numbers all, I, I could say these numbers are so consistent, I would predict it in any hospital added, and I stand by those words, that if you use AI, the proper AI used correctly, you'll see an economic lift, and I'll stand by that, that it increases an economic lift at the practise because we've repeated this again and again and again. The fact, the factor that we haven't calculated is actually the additional, additional procedures, diagnostics and treatments that result from an a revisiting of, of medicine, a revisiting of films and medicine. And there's actually a significant increase in there, but I can't assign a number to it.
So we're gonna, we're going to leave, leave that out. That said, there is consistently an increase in, in additional medicine once we start investing more in, in medicine. Next time, please pass.
Well, that's the, that's the next slide that consistently there is a revisiting of diagnostics and it may be an abdominal ultrasound is recommended when one wasn't before. And the increasing treatments often you pick up arthritis that, that in close questioning the owner, whether it's spondylosis and they're not jumping on the bed, or whether it's a significant arthritis of the stifle or the shoulder or the elbow. And close questioning that the, the patient is getting stiffer and it, it's appropriate to start looking at pharmaceutical intervention for that.
We, we see that or additional procedures where there's more confidence to do the diagnostics, often in the abdomen, we start finding more, more pathology and have the ability to address that pathology earlier rather than later. Next case, please. Next slide.
So what, what is the future? The future for AI in veterinary medicine is one, I see continual improvements of the systems available. It's a, it's a new field and it's expanding very rapidly.
The, the benefit is definitively there, the quality is definitively there, and at the same time, it's expanding and increasing. And so the next, the next several years, I would anticipate it's going to be really, really exciting. As AI expands into, into further areas of medicine, I would, I would, on one hand, say, yes, in radiology, there will be expansion of the tests.
At the same time, we're now seeing AI and cytology. My anticipation is there'll be AI in other areas such as interpretation of lab work, and then looking at combining different data sets. So radiology is a data set.
Laboratory is a data set, say physical exam is a data set, and history is a data set data set. There's no reason not to use AI in combining all those data sets to give us very strong predictive suggestions of, of next steps to next steps to take in a case. Yeah, the final thing is the efficiency factor.
By all means, this empowers veterinarians, and the example that I would use is the power suit. So, sorry, is the movie Iron Man, where, if anybody's seen the movie, I forgot, I forgot the actor's name, Downey Junior, what's his name? But anyway, he puts on the iron suit.
He's not replaced by the technology, but now he is, he is stronger. He's able to fly, he's able to shoot, he's able to be more effective. He's able to get the bad guys.
He's able to rescue the people that need to be rescued. As a veterinarian, I see this as simply very straightforward. Robert Downey Jr.
Somebody who said thank you, simply empowering veterinarians. I see the healthcare teams, the veterinary nurses or technicians becoming empowered and again further empowering the veterinarians. I don't see this as a challenge.
I'll state very strongly, a strong belief and a strong, what I've seen is it's an empowerment and it's an ability for us to get to move or move forward. It's not challenging anybody for their possessions at all. And I, and I, and I feel very confident about that.
Next slide, please, Ma. Just to point, some of the, some of the areas that are next is now that, now that we can break down radiology into consistent assessments and, and there is a, there is a standard approach to radiology. We can actually go into different areas such as develop a chart, and this is a chart we have.
It's so big we had to highlight some of the things. It almost covers the whole wall, but it's made available for, for hospital teams where you can have in the dark blue, you see, for example, vertebra heart score, identify the essential views. You don't always have to take, say, 3 views of a In order to, to get to, to be diagnostic.
And we're finding that more and more commonly identify the central views to, to, to make the diagnosis, what AI needs, what, what it's defined by, the common differentials, and helpful, helpful next steps, additional help next steps. And this, and this is not just for veterinarians, it's actually designed for healthcare teams to make better recommendations to the veterinarians that they are working with. Next, next slide please.
And that's it. So again, where are we on time? Oh perfect.
We're doing pretty well. No, thank you so much for that. That was so interesting.
Absolute pleasure to be chairing this because I'm learning so much, and we have a few questions and if there's any more guys, keep, keep them coming, in the Q&A box, but, if I, if I can sort of start, start asking you some questions from the delegates, so somebody's asked very, very early on just just as you were starting, about why does it not work with the skull radiographs. Ah Someone is paying attention. All right.
We haven't created the algorithms for it yet. Simply said, it, it would work. It, it comfortably would work.
We went for the most common, most common, first. So most commonly, the thorax, the abdomen, and the, and the, and the major parts of the limbs and the spine, and we simply haven't, simply said we haven't done, done the skull yet. And is that, is that sort of, are you, are you expanding into the I, I, I do.
We actually had a, we do the skull or should we do dental? And since dental films are becoming much more popular and folks often are switching from taking skull films, look at the teeth and to to actual dental films, we decided to do dental films and then we'll have to come back and do the skull. And then, and then there's other areas right now.
Just focused on dogs and cats. Certainly, birds are, are come comes up a lot, question about doing birds and then other things such as, horses and especially distal horse limbs. For pre-purchase exams and such, that that would really adapt to AI simply said haven't haven't done it yet.
No, it's it's great to hear and I spent a lot of time doing horse vet work, so, it's, it's good to hear. Yeah, no, absolutely, and, and we had a few other questions about sort of what other species and exotics and so on, so you've just. OK.
Answered that for us as well and there was a a very interesting question here I found, would be, I'm, I'm sure a lot of people who would want to know the answer to that is what, what does it really entail to set the system up, the single pet in a first opinion practise or is there the software installation or some like how, how, how do you set it up, you know, if somebody's interested. The, and so I, I, I try to avoid the commercial side of things, so I, I will answer, the, the, the beauty is most radiograph machines, most X-ray machines are connected to the internet. And as long as the machine is connected to the internet or the PA system is connected to the internet, we actually have technicians that with the approval of the hospital, who connect behind the scenes.
So it doesn't require a visit. There's no hardware, and it actually just takes a couple of minutes. And then, so then what happens is as the, as the radiograph is taken and obtained, the processor sends an image to us in the cloud, just as well as an image that is sent to the local storage as well.
And so we receive the images in the cloud just as the films are taken, and they can process it very quickly and turn it around and make, make the answers available. So it takes a few minutes to set up. Yeah.
Yeah, and, and actually just following up on that, somebody's just message that they've been using Singopet for about 6 months and they were asking how do they get the charts, and yeah. If you can let us know, happy to send you. In the next room, we've got a whole bunch, whole bunch of them.
So, happy, happy to, happy to send you a chart. Yes, perfect, thank you. It was a very simple question, but we, I, I thought, you know, it's, it's going to be helpful, so, .
Perfect. And there's quite a few questions actually, coming up with, a few different people have asked about whether you're planning to, marry AI with CT scans, MRIs, or, you know, you mentioned a few of the different sort of, areas, so, you know, where are we with that? It's a great, yeah, it's, it's a great question.
And actually, we do intend to do so with both CTs and MRs, and it's particularly useful because there are a lot of images for the cross-sectional scanning compared to say 3 or 4. And at the same time, there's actually greater consistency in the MRs and CTs given the window that, that those modalities look, look at the, at, at, at, at the subject or the patient. And so it's actually pretty, pretty straightforward.
In human medicine, a lot has been done with CTs and MRs because they're used so, so commonly. We started with plain films just because that is most common in veterinary medicine, and I would anticipate we will move quickly towards, cross-sectional imaging, as well. But we, we identified what would, what would give the greatest impact and the greatest impact would be plain films because most practises still rely on.
Paint films as the first modality, yeah. Perfect, fantastic. Thank you.
. And I have another sort of couple questions that are I guess similar style but slightly different, so I'll just ask the first one, they were asking about an anonymous question whether the AI takes into consideration or breed specifications, like for example when you're looking at virtual heart scores that may, you know, deviate from sort of normal based and on the, on, on breed? Great question. Great question.
If, if you look at VHS, VHS is a good guesstimate of heart size, in the long axis, the short axis of the heart or in a perpendicular compared to typically, the, the, the length of the of the 4th thoracic vertebra in body. And it's, it's a good estimate. It's not 100% perfect for two reasons.
One, to point to the question, there's breed variation, and two, it actually can change a little bit in the respiratory cycle of the patient. So the, the heart size can change a little bit, in appearance, and, and it's typically not tangible, but that change is the delta is, is there. There have been some papers published on breed size comparing, say, the more the anatomy of a, of a Yorkie versus a greyhound versus a German shepherd versus a brachycephalic.
Such as a pug or a, or an American bulldog or such. And they, they, there are some variations. If you look at, if you look at those papers, they actually they have very few numbers.
It's, maybe a couple of dozen. Examples, maybe 100 examples. It was several years ago that I looked at it, at least at that point.
They, they weren't enough to differentiate. We, we actually have millions and so right now we lump them together. Our next step will be to separate that out.
So this is where academia would, would be particularly helpful. And actually publish, publish the data that we have because we have millions of cases that can actually truly pinpoint, what are the actual breed sizes. There aren't, so the point is there aren't breed normals now.
We give a general, general heading. It covers the vast, vast majority of cases. I'm comfortable with that.
Over time, we will separate that out to different breeds. It'd be a great project for academia to separate that out. Yeah.
And we, we have, again, we'll have millions of cases to use. Fabulous, thank you. And I have a few other sir I guess more technical vetty questions coming through.
So a very relatively quick one. Could the AI be applied to hip and elbow scoring, do you think? Very easily, yes.
And it actually, we, we, we do, we do it for both hip and elbows currently. And, we don't, the only thing we don't do is look for, say, the growth abnormalities of the elbows and you're not anal process or such. And so I would anticipate we will do that again, expansion of tests we'll do that.
Those are less common and we'll do those over time. But looking for arthritis, looking for changes, certainly looking for arthritis of the hips or malformation of the hips, that, that, that's, that's more common and we actually, we actually have a panel for that. Yeah.
Perfect, fantastic and sort of still on that, technical topic is, there's a couple of questions that popped up about, how, how important is this positioning and exposure, you know, during this interpretation and also is there a way that the images get. The, the vet can choose which images are, evaluated or does it evaluate all of them? Great, great.
And there's one more question I would add, add to that. Let me explain. So great question.
One is, does it have to be a perfect exposure or technique? Of course, it's ideal, ideal to be as close to perfect as possible. There's no, no question about that.
I don't want to get myself into trouble. At the same time, if we look at the, the litmus of With an average veterinarian look at those films and be able to interpret? And if an average veterinarian can interpret those films, even if they're a little turned, even if after the fact you have to adjust the exposure, then The AI definitively can make calls on that.
And, and that's, and that's a safe because the AI is actually a little bit more forgiving because it looks at the films differently. It looks at those patterns differently in, in, if you will, in a different dimension than we can with, with human eyes. And so it is, I, I don't want to get myself into trouble, but it is a little bit more forgiving than we've typically become used, become used to as far as, as far as technique.
The, the second question, the second question, sorry, what was the second question? It was about choosing which images are evaluated. It's, no, it actually, it's an automatic process.
It looks at all the images. At the same time, there's not a price per film. So.
Yeah, the question then continued actually asking sort of in case exposures have to be repeated. Does it, does it do every, you know, exposure is evaluated or it actually looks, it actually looks, and it's not typically a problem with the, the what we call whole body, but the thorax, the abdomen, the spine, the limbs. It's not typically a problem there.
But with dental films, you can get a lot and a lot of dental films. And so only when it It starts hitting a double-digit maximum on films, we tend to ask that those numbers decrease just a little bit, but typically it's not a problem looking at each exposure, it looks at, at each film for everything each time. That's, that's the benefit of, benefit of technology.
It's not a price on the body. There is a little bit of, of an additional price on dental, but just because it ties up the system, because there's one study can end up being lots and lots and lots and lots. Lots of films and only if that happens, we come back.
And there's one other item that we haven't made available yet is we do, we do look something to come. We do look at the quality ourselves. So even though we don't release it back, we do look at technique.
Is the animal, positioned correctly? Is the animal breathing at what point in the respiratory cycle, if it's a thoracic film, Is it too dark? Is it too light, All, all of those, all of those aspects we do actually collect data on, we do analyse, but we don't today, we don't report back on.
In the future, I anticipate reporting back live for the folks taking the films. So in other words, the veterinarian doesn't have to quality control whether these are good enough films or not, the system will do it for the folks who take the films, and that'll save the veterinarian an extra step of having to quality control or approve the films. Yeah.
Oh, fantastic. That's so interesting. And I have, I have another couple questions before we start wrapping up and one actually is quite an interesting question, particularly because you actually mentioned sort of potentially moving into pre-purchase exams for horses.
One of one of the questions was, whether these reports are valid in case there is any legal issue. How have, have you had any cases that you, yeah. It's a great question.
It's actually an excellent question. I'll speak primarily for the US, if you, if I will, and explain it. There hasn't, there hasn't been, if you will, a big legal, challenge.
Look forward at some point it'll happen, I think. And actually, I think what we will do is look at radiology in general. So for example, in the US, human radiologists need to be licenced in the state that they're providing a service.
So wherever the radiologist is, it doesn't matter. telemedicine certainly occurs, but in human medicine, wherever the patient is, the radiologist has to be licenced in that state. So if If a radiologist is in Hawaii and reading on a, on a human patient in Kansas, they have to be licenced in Kansas.
Veterinary medicine radiologists operate in a bit of a loophole, so they don't, they're not required to be licenced in, in every state that they need. It's just a loophole that exists, that eliminates that sort of liability for radiologists. That said, I think that that's appropriate.
I'm not saying that those rules should be toughened. But what happens is artificial intelligence at this point falls under the same rules. There's not a specific licence for artificial intelligence in the United States.
It falls under the same rules of, of, excuse me, of veterinary radiology by most boards and that they see it as a same sort of information. If there's a question, it still comes back to the practitioner. So whether it's a veterinary radiologist report or an AI radiologist report, if there's a question on the case, it is always 100% dependent upon the practitioner that has the relationship with the client and the relationship with the pet.
And, and most state boards, my communication with the state boards is that they'll fall into the same category as the radiologists that way. Fantastic. Thank you very much for that.
And I think we're kind of going a couple of minutes over time, but there maybe one last question to wrap up. It's quite a nice, hopefully quite a quick question. But somebody's asked about the accuracy, you know, what's the accuracy rate of AI obviously, you're using it, we're using it and it's growing, so, I assume it's pretty up there, but, I thought I'd let you, great question, great question.
There is, there, the accuracy and we can in, in broad terms, break down accuracy into sensitivity and specificity. So false positives, false negatives, and calculated and such. The, the accuracy of the AI is actually, we won't release it unless it's above 95%.
So our, each of the test panel will not, will not be released and, and, and we prove that accuracy and we have lots and lots and lots and lots and lots of data to demonstrate it as well as an ongoing, ongoing quantity control processes to assure it and further improve, improve, improve. And so we won't release it unless it's 95%. Will it get to 100%?
I don't think so. So what is the accuracy of a human eyes in human studies with human radiologists looking at human patients, the accuracy is around 80%. And veterinary in veterinary eyes, I haven't seen many studies that have had any significant numbers at all.
In on on the human side, they generally talk of an accuracy of about 80%. There is greater consistency with, with the AI systems and, and that's the benefit that we can apply, but to the point of the previous question, it's always up to the clinician to make, yes, I finally agree or I don't agree, and that's up to the clinicians see the patient. Yeah.
Perfect. Thank you very much for that. And I have a lot of thank yous coming through, superb talk webinar, sorry, great lecture back to work.
So thank you, thank you very much to Signal Pet for, you know, kindly sponsoring and thank you for being here to giving us this this talk. I don't know, do you have any last last words of wisdom before I wrap up? The, the last word is, it really is, it's exciting, it really is an exciting future for being literally being in medicine my, my entire life.
I see this as the new future. When something's new, it can feel threatening with confidence and speaking as a proud, proud veterinarian, it's nothing to be intimidated by. It's something to be harnessed and used to help our patients, to help our clients, and to help our hospitals and they can do so.
As we look forward, yeah. Fantastic, and I, I absolutely love innovation and improvement and you know, improving the processes. So I really enjoyed the webinar.
I see that a lot of the other delegates have really enjoyed, so thank you very much again. Thank you. I appreciate it.
Thank you. Time, yeah. Bye bye, bye bye.
