X-ray interpretation for small animals

[orainformatics]

Questions for people leaning toward small animal practice. Do you feel your radiology training is preparing you the way you like? Do you have strong feelings about using telerads for your cases? What would be some roadblocks to adopting an automatic AI interpretation? Either personal preferences or from a tech standpoint? Any other thoughts on AI in practice?

---

Members don't see this ad.

 

[DVMDream]

What is an automatic AI interpretation?

 

[WhtsThFrequency]

My first though would be the cost issue, as well as the lack of AI software (as far as I know) for veterinary patients.

1) Where is the money going to come from to develop AI interpretation software/algorithms for animals? Let alone multiple species (dogs, cats, and horses make up the majority of patients needing x-rays, but there is a lot of variability in those when you start accounting for breeds as well - a normal Dachshund's leg x-ray will look a lot different than a normal Labrador's leg x-ray, for example) and dozens of different organs/views? Getting funds for veterinary research is difficult anyways; I can't see many major funding organizations wanting to pour money into that. How would it be validated? How long would it take?

2) How much money would clinics have to invest to be able to take advantage of AI? Most clinics have a very tight profit margin.

3) How would this cost be reflected to the client?

Sending rads digitally to specialists for interpretation is common enough - so teleradiology isn't a big issue. I personally think it's great.

I felt decently comfortable with major radiologic changes in both small and large animal patients after school. I do wish we had gotten more ultrasound training, though.

 

[CalliopeDVM]

an automatic AI interpretation?

Do you mean that the rads would be read by AI every time they're taken? I'm personally pretty skeptical about AI in veterinary radiology because of the large variation in our patients (far more than in human medicine).......I'm keeping an open mind, but I'm far from ready now to accept it. I'm also not sure that practice owners would be willing to pay for any service to read all their rads, when at least half of the rads most GPs take are pretty easy to interpret (yes, there are bladder stones or rocks in the stomach or a twisted stomach, for instance). I can see AI as being of more use to radiologists rather than general practitioners.

 

[SocialStigma]

Not sure how I would feel about AI interpretation, part of what I like about teleradiology is the ability to contact the radiologist who read the images for further clarification if I need it. I read all of my own orthopedic imaging and abdominal/thoracic radiographs but send out all my surgical oncology case planning CT scans for radiologist report.

 

[Minnerbelle]

I would say **** no at this time. I send a majority of my rads out to a DACVR right now and they miss things all the time... I’m not sure I would trust AI at the moment.

 

[orainformatics]

My first though would be the cost issue, as well as the lack of AI software (as far as I know) for veterinary patients.

1) Where is the money going to come from to develop AI interpretation software/algorithms for animals? Let alone multiple species (dogs, cats, and horses make up the majority of patients needing x-rays, but there is a lot of variability in those when you start accounting for breeds as well - a normal Dachshund's leg x-ray will look a lot different than a normal Labrador's leg x-ray, for example) and dozens of different organs/views? Getting funds for veterinary research is difficult anyways; I can't see many major funding organizations wanting to pour money into that. How would it be validated? How long would it take?

2) How much money would clinics have to invest to be able to take advantage of AI? Most clinics have a very tight profit margin.

3) How would this cost be reflected to the client?

Sending rads digitally to specialists for interpretation is common enough - so teleradiology isn't a big issue. I personally think it's great.

I felt decently comfortable with major radiologic changes in both small and large animal patients after school. I do wish we had gotten more ultrasound training, though.

Hi and thanks for your thoughtful reply.

Regarding 2 and 3 - what if a certain percentage, say 25%, of the films you normally send to tele were interpreted with AI and confirmed your clinical suspicion and you only had to send the trickier cases for a tele read?

If you send 100 films to tele per month now at $80 per read, that would be $8000. If AI was priced at $100 per month and saved 25% ($2000) and you still sent 75 films, would that be attractive for a veterinary practice?

You bill the client for radiographs like usual with a small percent upcharge for AI service, $1 in the example above and if you send for tele, you bill for that as usual.

In another way, you read many of your own films and send out the ones you'd like another opinion. Is it feasible an algorithm could help with something specific like heart failure or subtle tracheal collapse and decrease the number of the radiographs you may normally send out for tele? Maybe save money for the client?

Thanks again for your reply.

 

[orainformatics]

I would say **** no at this time. I send a majority of my rads out to a DACVR right now and they miss things all the time... I’m not sure I would trust AI at the moment.

Thank you for your reply.

What if the AI was a screening layer and you still do your own eval and send out your usual films. If it was priced reasonably, is that worth a try?

Thanks again.

 

[orainformatics]

Not sure how I would feel about AI interpretation, part of what I like about teleradiology is the ability to contact the radiologist who read the images for further clarification if I need it. I read all of my own orthopedic imaging and abdominal/thoracic radiographs but send out all my surgical oncology case planning CT scans for radiologist report.

Thank you for your reply.

Concur, oncologic case planning is outside the scope of any AI medical imaging platform at this time.

 

[orainformatics]

Do you mean that the rads would be read by AI every time they're taken? I'm personally pretty skeptical about AI in veterinary radiology because of the large variation in our patients (far more than in human medicine).......I'm keeping an open mind, but I'm far from ready now to accept it. I'm also not sure that practice owners would be willing to pay for any service to read all their rads, when at least half of the rads most GPs take are pretty easy to interpret (yes, there are bladder stones or rocks in the stomach or a twisted stomach, for instance). I can see AI as being of more use to radiologists rather than general practitioners.

Thank you for your reply.

That is a very good point, where is the boundary of what a GP is 100% comfortable reading? And when it is sent to tele?

Very likely doctor dependent.

If you send half as an example and that costs $5000 per month. What if AI could shave 10% off of what was sent to tele because you grew more comfortable with a slightly trickier finding? Would that be helpful to the pet owner?

Thanks again for your reply.

 

[Minnerbelle]

Thank you for your reply.

What if the AI was a screening layer and you still do your own eval and send out your usual films. If it was priced reasonably, is that worth a try?

Thanks again.

probably not. If I’m hiring a radiologist, I don’t feel the need to send another screening layer. If I‘m not sending to a radiologist, I’m comfortable with my own interpretation.

 

[dyachei]

What minnerbelle said. I would likely not be comfortable with an AI at this point. There are so many variations in radiographs, particularly when breeds can be so conformationally different.

 

[DVMDream]

I wouldn't trust an AI model in veterinary medicine. It would require a lot of research, planning and scientific studies proving that AI interpretation is accurate and there's no way a company is going to want to dump the amount of money needed to do those things that may persuade me that AI could be a possible option.

 

[WhtsThFrequency]

What minnerbelle said. I would likely not be comfortable with an AI at this point. There are so many variations in radiographs, particularly when breeds can be so conformationally different.

I think that’s the key here. What is abnormal in one breed may be normal in another. Sure AI might be able to dx heart failure or collapsing tracheas, but any GP should be able to do that on their own.

 

[CalliopeDVM]

what if a certain percentage, say 25%, of the films you normally send to tele were interpreted with AI and confirmed your clinical suspicion and you only had to send the trickier cases for a tele read?

That can be a self-fulfilling prophesy, though, and not always in a good way. Would a vet be more willing to accept the AI results that confirm her suspicion and therefore not send it off to a radiologist, when in fact the AI and her suspicion were both wrong? How would a vet know which ones are "tricky" and which ones aren't? As Minnerbelle said, do vets need a middle layer of interpretation - additional to their own reading but not a Radiologist's reading? I don't think so.

That is a very good point, where is the boundary of what a GP is 100% comfortable reading? And when it is sent to tele?

Very likely doctor dependent.

And location dependent, and client dependent; it's about far more than just whether or not the GP is 100% comfortable. Some locations are more likely to immediately refer some potentially complex cases (eg HBC) where rads will be checking for multiple possible systems problems (so won't take and need to read those rads), whereas others have no referral options nearby and will need to tackle them themselves (and would use tele rad services). And some clients (often also location dependent) will have funds too limited to allow for any additional costs......even if the GP isn't 100% comfortable, sometimes they have no choice but to do it anyway. And then, yes, it can be doctor dependent.

 

[WhtsThFrequency]

Thank you for your reply.

That is a very good point, where is the boundary of what a GP is 100% comfortable reading? And when it is sent to tele?

Very likely doctor dependent.

If you send half as an example and that costs $5000 per month. What if AI could shave 10% off of what was sent to tele because you grew more comfortable with a slightly trickier finding? Would that be helpful to the pet owner?

Thanks again for your reply.

This would all depend on the accuracy of AI for the wide variety of patients we see. Which is what is really in question here...so I don’t know if we can really give you a good answer.

 

[CalliopeDVM]

What is abnormal in one breed may be normal in another. Sure AI might be able to dx heart failure or collapsing tracheas, but any GP should be able to do that on their own.

Exactly - what appears to be cardiomegaly might be normal for that breed. Of course, AI could be taught all that stuff -- variations in breed and age-related artifacts -- but I'd want to make sure that's a part of what's being proposed.

 

[battie]

Exactly - what appears to be cardiomegaly might be normal for that breed. Of course, AI could be taught all that stuff -- variations in breed and age-related artifacts -- but I'd want to make sure that's a part of what's being proposed.

And then you'd have to consider mixes. For a lot of owners what their mixed dog actually is really anyone's guess. So then what could be normal for that dog's mom might still be abnormal for that dog

 

[JaynaAli]

I agree with these comments, plus what about quality issues with rads? It's not like a human that will lay there completely still as long as you need. Obliqued images, artifacts, expiratory films...those are all things that people deal with every day in vet med that would have to be accounted for in an AI setting and make it much harder to implement even if the money was there.

 

[WhtsThFrequency]

I agree with these comments, plus what about quality issues with rads? It's not like a human that will lay there completely still as long as you need. Obliqued images, artifacts, expiratory films...those are all things that people deal with every day in vet med that would have to be accounted for in an AI setting and make it much harder to implement even if the money was there.

That's another good point. Sedation and holding are always imperfect. We can't just tell them to be still.