Latest / CattleUSA Daily / 255: Could AI Predict Heart Failure Risk in Cattle? with Chase Markel

Transcript

255: Could AI Predict Heart Failure Risk in Cattle? with Chase Markel · CattleUSA Daily

0:00
In this episode of the Cattle USA Daily podcast, we are
0:03
welcoming on Chase Markel to the podcast.
0:06
Chase is a student at the University of Wyoming pursuing
0:10
his PhD, and Chase has been doing some absolutely
0:15
outstanding research work within the cattle industry.
0:19
Chase is here today to talk about everything that he has
0:23
been doing is learning and some early findings that they have
0:28
kind of found from their research that they have been
0:31
doing over the past few years when it comes to everything that
0:36
he is doing within the industry. I'm not going to get into it
0:39
now. Chase will explain it all here
0:41
in just a few seconds. For you guys, if you're watching
0:44
on YouTube, make sure you take a second to hit subscribe and like
0:47
this episode. And if you were watching on any
0:49
of our other podcast platforms, make sure you take a second to
0:52
make sure you're following the show and rate and review as
0:54
well. Without further ado, let's go
0:57
ahead and hear from Chase. Welcome to the Cattle USA Daily
1:04
Podcast, your go to source for the latest livestock markets,
1:07
industry insights and the updates shaping agriculture
1:10
today. I'm your host Lauren Moylan, a
1:13
passionate advocate for agriculture, here to keep you
1:15
informed and connected. Now let's dive into today's
1:19
episode. Chase, welcome to the podcast.
1:22
We are so excited to have you on.
1:24
When I first read about the research that you are doing at
1:28
the University of Wyoming, I was just super excited for what
1:32
you're offering the industry, what you're doing, and just the
1:35
fascinating side of something I personally probably would have
1:40
never thought about is even a concern within the cattle
1:43
industry. But I would love it if you could
1:44
just take it away, share a little bit about yourself, your
1:46
background, and then the research that you're doing and
1:48
kind of what you're doing right now.
1:50
Yeah, Well, that thanks, Lauren, for having me.
1:52
Like you said, my name is Chase Markel.
1:54
I grew up in Wheatland, WY I've kind of been a ranching
1:57
background, grew up around cattle my whole life.
1:59
So yeah, that's something that I always wanted to do and it's
2:03
still pursuing it to this day. But I kind of bounced around
2:06
school to school. I was on livestock judging
2:08
teams, so I judged in junior college at Northeastern Junior
2:12
College for a few years before transferring to the University
2:15
of Wyoming, where I'm at now. So I finished my undergrad from
2:18
at the University of Wyoming and then just hopped right into grad
2:21
school, did a master's program here.
2:23
And that's kind of what kicked off this, this area of research
2:28
and what we're going to talk about today.
2:30
I started doing research on brisket disease and bovine
2:33
congestive heart failure. You know, that was something
2:35
that I kind of knew about from a 30,000 foot view, but I didn't
2:38
know the intricacies. And then fortunately in my
2:42
graduate program, I was able to dig into that and really fell in
2:45
love with the topic and the problem at hand.
2:48
So I got to study that for two years and then I'm a glutton for
2:51
punishment. So I stayed around for a PhD and
2:54
I'm still doing the same thing. So along the way I kind of got
2:58
interested in in mathematics and statistics.
3:01
I'm kind of a nerd in that way. So I took a whole lot of stats
3:04
courses, got a minor in statistics, and I started to
3:07
realize how important that is for this industry because the
3:12
beef industry is so data-driven and in agriculture in general is
3:16
data-driven. And it's extremely important to
3:20
know what to do with that data. So I got I really obsessed with
3:23
statistics and, and trying to figure out how can I apply this
3:26
to problems that can help Someone Like You or me or, or my
3:30
neighbor down the road. And that's really what I'm
3:33
trying to do right now is use data science and math to to help
3:38
us make better decisions and hopefully make the the industry
3:42
more efficient. You are truly doing the work
3:45
that none of us really want to do.
3:49
If we're on a Cal Cal operation every single day, we don't, we
3:52
don't want to be sitting down and doing the math and figuring
3:54
all the science and the research and all of those things.
3:57
But so I thank you for your work and what you're doing to further
4:00
the industry. But before we dive too far into
4:03
the topic, can you talk about just why you started to get into
4:07
the pulmonary hypertension and what that means for the cattle
4:10
industry and kind of how that relates to our producers today?
4:16
Yeah. So I kind of just got thrown
4:18
into a project and they said we're doing the speed lot
4:21
project, we're going to look at bovine congestive heart failure.
4:24
And at that point I, like I said, I didn't really understand
4:27
what it was. And then I, fortunately,
4:30
throughout the years, have really dug into the topic, got
4:32
to work with some of the greats in this subject, Doctor Tim Holt
4:36
being one of them. He's a great mentor of mine.
4:38
And it's a really important issue, especially where I'm at
4:41
in Laramie, WY. We're at 7200 feet in elevation
4:45
and brisket disease is something that every producer in Laramie
4:49
pays attention to. They have to so at first, you
4:53
know, started to study brisket disease here at the university
4:56
and, and it's a really important disease.
4:58
You know, people that are at lower elevations have maybe
5:01
heard about it and heard that, you know, there's a certain
5:03
bulls that we should use, we shouldn't use.
5:05
But it can be extremely devastating.
5:09
You know, there's reports every year of, you know, 10 to 15% of
5:13
a calf crop being lost if you know, something goes wrong with
5:17
those genetics. So it can be a major problem and
5:21
really what it comes down to, you know, I mentioned elevation,
5:23
but it's just an oxygen problem is all it is.
5:26
So like you mentioned, pulmonary hypertension is what causes
5:29
brisket disease. So that is the blood pressure in
5:33
between the heart and the lungs. So when there's a lack of
5:37
oxygen, we call it hypoxia, and that lack of oxygen causes those
5:41
small vessels in the lungs to constrict.
5:44
So it's trying to push that blood to areas of the lungs that
5:48
are ventilating more efficiently, so getting more
5:50
oxygen. And that's a really great thing
5:53
that that these bodies that their body can do, but
5:55
unfortunately, after long periods of time, it starts to
5:59
make them sick. So those vessels now they're
6:01
constricted for a short period of time and after a while, if
6:04
that hypoxia isn't resolved, then those vessels will actually
6:08
start to hypertrophy. So they'll get thicker and
6:10
thicker and they'll start to harden and then that pressure is
6:13
stuck at that elevated rate. Once those vessels begin that
6:16
process, all of this pressure is backed up into the right side of
6:20
the heart and the right ventricle.
6:21
And so now that the right side of the heart, once this elevated
6:24
pressure reaches that threshold, is working double time and
6:28
eventually if something doesn't resolve itself, that animal will
6:31
likely die of congestive heart failure.
6:34
So the right side of its heart essentially just gives out.
6:37
So that's what causes brisket disease.
6:39
Elevation is the hypoxic stressor is what I like to call
6:43
it for brisket disease. You know, when you go up in
6:46
elevation, there's a decrease in atmospheric pressure.
6:51
That decrease in atmospheric pressure causes those oxygen
6:53
particles in the air to get further and further apart.
6:56
So those animals are getting fewer oxygen particles per
6:59
respiration event and that can lead to this hypoxic cascade.
7:03
Now, I also mentioned bovine congestive heart failure.
7:05
And you know, there's this conversation in the industry of
7:08
are these two things the same thing?
7:10
Are they different? What's the difference?
7:12
And I can tell you it's a little bit of both.
7:14
So obviously at low elevation hypoxia in the way that I
7:18
described it isn't necessarily a problem.
7:21
So there's got to be some other type of hypoxic stressor.
7:24
And right now, it's not really well understood what causes that
7:28
hypoxic stressor. It really could be a whole lot
7:31
of different things. It could be respiratory disease,
7:34
could be a dust pneumonia, it could be an increased need for
7:38
oxygen because of higher metabolism and feedlots.
7:41
I mean, it could be all kinds of different things.
7:43
So we really haven't been able to pinpoint exactly what is
7:46
causing that hypoxic event. But what I can tell you is the
7:50
cascade afterwards in the pulmonary system is the same.
7:53
You'll see that constriction of the smaller vessels in the back
7:56
of the pressure into the right side of the heart and those
7:59
animals will succumb to a congestive heart failure.
8:01
This process in feed lot animals generally happens when they're
8:04
quite a bit older towards the end of the feeding period.
8:07
Whereas brisket disease is often, you know, right around
8:10
yearling age or calves, you know, in summer pasture will
8:14
will die from this disease. So it happens at different time
8:16
points, it happens at different production cycles, but as far as
8:21
the pathophysiology goes, it's essentially the same after that
8:25
hypoxic stressor. Yeah.
8:27
Can you talk a little bit more about the significance of that
8:29
right ventricle and the risk for that pulmonary hypertension and
8:33
how that kind of like leads to heart failure from there?
8:37
Yeah. So the, the right ventricle is
8:39
extremely important. It is the thing that feeds the
8:42
blood to the lungs. So if your right ventricle isn't
8:45
working, your your blood will not get any oxygen and then the
8:49
tissues that need that oxygen are being deprived of oxygen.
8:52
So it really is the Tran, this transport method of that of that
8:57
blood into the pulmonary system. And obviously over time, if the
9:02
tissues in the body are not getting enough oxygen because
9:05
that right ventricle is failing, it's, you know, it's maybe not
9:08
pushing the volume of blood that's necessary.
9:10
We can see all kinds of issues. And we see this in comorbidities
9:15
with congestive heart failure. You'll start to see liver
9:17
congestion. So they'll have passive, passive
9:19
liver congestion because of a venous drainage issue, because
9:23
of the failure of that left venture or the right ventricle.
9:26
Excuse me. And systemically, there's a
9:28
whole lot of problems. You will see lymphatic tissue or
9:32
let lymphatic fluid start to pool in different tissues.
9:35
And then the main thing and the reason why you know, brisket
9:38
disease got its name is because of this, this hydrostatic
9:42
pressure is what it's called in that pulmonary system.
9:45
An aqueous kind of protein filled fluid leaks from the
9:49
circulation and it'll fill in that brisket area.
9:52
So they'll get that swelling and that fluid buildup in the
9:55
brisket. And, you know, that's how that
9:58
was was named. So the right ventricle is really
10:01
kind of the crux of all of this. And as soon as that fails, that
10:04
animal can get very, very sick in a lot of different ways.
10:08
Yeah, absolutely. And it's really significant,
10:10
especially when we're talking 1/3 of a calf crop like that
10:13
just presents such an economic impact to many ranches,
10:17
especially in those higher elevations where this is a very
10:19
real reality of your guys's ranching lifestyle.
10:23
You built an AI model that kind of helps you analyze a lot of
10:27
different aspects of your research.
10:29
Can you talk a little bit more about that and what you've done
10:31
there? Yeah.
10:32
So one of the projects that I'm working on and kind of what
10:37
we're going to talk about today is focused on bovine congestive
10:40
heart failure, so not necessarily brisket diseases or
10:43
feed lot cattle. And what I'm especially
10:46
interested in is subclinical cases.
10:49
So when you hear about congestive heart failure, you
10:51
always hear about death loss in the feed lot.
10:53
And that's certainly very costly because it happened so late in
10:57
the production cycle. A lot of the data that we have
11:00
shows that, you know, that might just be the tip of the iceberg.
11:04
It works on this continuation. So an animal doesn't just go
11:07
from being completely healthy to deathly ill.
11:09
It's kind of this long drawn out process.
11:12
So there's a lot of animals that are in feedlots that actually
11:15
make it to the harvest facility that are in, you know, a certain
11:20
stage of heart failure. So I'm very interested in those
11:22
cases, mostly because they're really hard to diagnose.
11:27
It becomes easy to diagnose when we talk about the fluid buildup
11:30
and all these other things. And that's easy to diagnose, but
11:32
that's towards the end. And there's not a whole lot you
11:35
can do with that animal. But there is a, you know, a
11:37
degree of pulmonary congestive heart failure to where if we
11:41
catch it soon enough, there's some things that we can do in
11:44
order to prevent a complete loss.
11:46
And one way that we use to diagnose at the subclinical
11:50
cases post mortem is by heart scoring.
11:54
And I mentioned Tim Holtz a little bit earlier at Colorado
11:56
State University. He developed this system of a
11:59
ballot evaluating hearts on the off ball line and grading them
12:02
one through 5 on this continuum, one being a completely normal
12:06
heart and five being a severely affected, a congestive heart
12:11
failure case. And so we sit on the line and
12:14
evaluate these heart scores to try to determine, OK, what is
12:17
the prevalence of these subcritical cases?
12:20
And especially what I'm interested in is what production
12:23
variables are associated with those subcritical conditions.
12:27
They have lower gain, are they, you know, lower hot carcass
12:30
weight is there, you know, lower marbling scores, things like
12:33
this that, you know, I kind of call the silent killers.
12:37
You know, you may not may not notice them, but you know, when
12:40
you get the check from the plant, so it may impact you
12:43
there. So that's something that I'm
12:44
really interested in. So we sit in these these packing
12:46
plants and score hearts as they go.
12:49
And I don't know if you've ever been on a kill floor of a
12:52
packing plant, but it's not the most glamorous place in the
12:54
world. It's kind of hot and sweaty and
12:56
you're sitting there all day and there's a million things going
12:58
on and you're trying to score hearts and sitting there for
13:01
hours on end. And I just got to thinking, man,
13:03
there's got to be a better way to do this.
13:05
And I made some connections previously with some people that
13:08
were doing computer vision and machine learning, specifically a
13:12
faculty in the anthropology department of all places at the
13:15
university, using a satellite images to evaluate ancient
13:20
structures. I mean, I just thought, well, if
13:21
you can, if you can find an ancient structure with a
13:24
satellite image, surely we can score a heart with an image of
13:27
of these hearts on the off ball table.
13:30
So I just started taking these images and started to collect
13:33
thousands of heart images. And I what I was able to do is
13:37
instead of on the line, trying to frantically score them and,
13:40
and be as accurate as possible, but keep track of where you're
13:43
at on the line. I was able to take those images
13:46
in sequence, go back, go back through, make sure everything
13:50
was in the right order, very carefully score those hearts and
13:53
make sure they got the correct score, and then train the
13:55
computer model based on, you know, that training data.
13:59
So I use the. It's a convolutional neural
14:02
network that evaluates these images and based on the training
14:07
data of previously classified arts, it learns from the way
14:11
that I classified the training data and automatically adjusts
14:15
itself to try to make it as accurate to my scores as
14:20
possible. And at this point, the model has
14:24
worked really well. You know, we kind of got lucky.
14:26
I'm not really a tech guy or anything like that, to be quite
14:30
honest with you. We didn't have to try very hard.
14:32
It just kind of worked right away.
14:34
And you know, this model is scoring hearts right now.
14:37
It's about 92% accurate. And with computer vision, that's
14:41
about as accurate as you can get it just because there's so many
14:44
variables. So now instead of graduate
14:46
students sitting on the line all day trying to score these
14:49
hearts, we can just start banking images and we can score
14:52
thousands and thousands of hearts.
14:55
And it's more efficient. But, but it's, it's more
14:57
consistent is the main thing. Because I might score a heart,
15:01
you know, I might call one a three where you would call it a
15:04
2. And you know, that can make data
15:06
kind of messy. But this model, it's perfectly
15:08
consistent every time. So now we have a perfectly
15:11
consistent set of phenotypes that not only can we evaluate
15:15
prevalence in any production, you know, anywhere we're at, but
15:19
we can also associate those scores with these production
15:22
variables, you know, live animal variables and carcass variables
15:26
to try to determine what this is costing the industry and and
15:31
then make adjustments to those cattle based on that.
15:33
So one thing that really excites me about this, and I think
15:37
there's maybe going to be this shift to sort of digital
15:40
phenotyping is instead of, you know, for like say for a breed
15:43
association or something like that, instead of subjectively
15:46
scoring a phenotype and submitting it to the breed
15:48
association can use a tool like this that scores everybody's the
15:52
same. So there takes the subjectivity
15:55
out of it and then we could build better selection tools and
15:59
improve the cattle that. I think it's really exciting to
16:03
hear about something AI can just see how passionate you are about
16:06
this chase, which is so exciting from someone in the industry to
16:10
hear about how we can move forward and how we can grow with
16:14
technology and use things like this that are very subjective of
16:18
like, yes, we have a guidance on it of how to score in certain
16:22
areas, but you said it perfectly.
16:24
You may score it a three and I may score it a 2 based off of
16:27
certain conditions in certain areas.
16:29
And so I think that you make such a great point on the larger
16:33
standing of what this can mean for the industry as a whole
16:37
rather than it's awesome for your research, right?
16:39
It's awesome for other research institutions.
16:40
That's that's great. But what does this mean for our
16:43
industry as a whole? And you made such a great point
16:45
there as far as really looking forward to this being something
16:49
that we can use across the board, especially for a lot of
16:52
our breed associations where like you said, a lot of that
16:55
phenotyping we are doing based off of a subjective matter.
16:59
And it's not providing the consistency that we need across
17:02
the industry to then have that data that is consistent and can
17:07
actually be relied upon. Yeah.
17:10
And another thing that I get really excited about when we
17:13
talk about this and it's sort of a hot button topic, but this is
17:17
great for building selection type tools.
17:20
But another thing that really excites me here is traceability.
17:24
So this captures a snapshot of, you know, in this case a heart,
17:28
but it could be whatever you, you, you want and ties it to
17:33
that animal, whether it's an Eid or an eye tag or ear tag,
17:36
something like that. So, you know, if something does
17:39
go wrong or we need to, to try to figure out to improve
17:43
something very specific, we all of a sudden we have this ability
17:47
to go back and visually look at it and then evaluate, OK, what
17:51
did the model say? What does it look like?
17:53
And I think it has huge applications, more so on the,
17:56
you know, aside from the breed association thing on, on the Fed
17:59
cattle side and the commercial side, traceability is a is a big
18:04
thing and I don't want to get into it too much, but I think
18:07
it's going to happen regardless. And I think this is just one
18:10
tool to, you know, instead of just referencing a data sheet
18:15
of, of what, what happened, we act like we have an image and we
18:19
have the, you know, the whole process.
18:21
So I think there's there's some applications there as well.
18:25
Absolutely. And I think it's when you look
18:28
at it in just a ability to pull data in a next level that we've
18:35
never seen before. I think that that is going to be
18:37
what sets us apart and really over the next 5 to 10 years,
18:41
puts our industry forward because we're going to be able
18:44
to gather data at another level than we've ever seen before.
18:48
Way quicker, way more efficient and at a higher accuracy level.
18:52
To then be able to take to our research institutions or to our
18:55
different associations and really see how different
18:59
decisions that we are making are impacting our industry as a
19:02
whole. When we're looking at your
19:04
research and what you're really trying to find, is there any
19:07
preliminary findings that you kind of have, anything that you
19:10
can kind of discuss there that you find exciting or just part
19:13
of your research? Yeah, Yeah.
19:16
Thank you for asking. Yeah, we've got a lot of things
19:18
going on and there's a lot of cool data that this spring, I
19:21
hope we're going to get quite a bit of it out.
19:23
But like I mentioned, these subclinical cases is what I
19:26
focus on, and it's maybe worse than what I thought it was.
19:30
So that's not necessarily exciting, but it's good that
19:32
we're finally identifying this. And when we talk about death
19:36
loss from bovine congestive heart failure, the acceptable
19:39
death loss rate is about .15%. So it doesn't make up that many
19:44
cattle. It's very costly just because of
19:46
how many resources are in those cattle.
19:48
But it's not this hugely endemic type of disease.
19:53
But when we look at subclinical cases, especially when we get to
19:56
a heart score, 4:00 and 5:00, so those would be severely affected
20:00
by by this disease. They make up anywhere from 4 to
20:05
7% of cattle. So you know, that's multiple of
20:12
what we see in death loss. And like I said earlier, if we
20:15
evaluate these production variables and there's a lot of
20:18
other institutions that are working on this as well.
20:21
But if we start to look at game, we start to look at hot carcass
20:25
weight, we look at marbling scores, things like that.
20:28
These cattle are sick. So you know, a sick animal is
20:31
not a very productive animal. And the data shows us exactly
20:34
that. And we've got some data and
20:37
certain data sets. It's, you know, 40 lbs lighter
20:41
in hot carcass weight in 10 cent increase in cost to gain.
20:45
So on the live animal side, it costs $0.10 more a day to feed
20:48
those animals just because they're not as efficient.
20:52
I mean, there's a couple of other things, but you just when
20:54
you think about this, these sub critical cases, they're just
20:57
absolutely hammering your efficiency and you know, you're
21:01
losing weight, you're losing efficiency in the feed yard.
21:04
And you know, I have some estimates that we haven't really
21:07
put out there yet, but I think it's reasonable to say that
21:10
this, this disease alone just in loss of, of efficiency is
21:14
costing the industry around $100 million a year.
21:18
So that's a pretty significant chunk of change that is very
21:22
troublesome because these are hard to diagnose.
21:25
So yeah, I mean, it's it's a big problem.
21:27
These subclinical cases are, are way more severe than I think
21:30
people realize. And you know, it's something
21:33
that I think the industry needs to focus on.
21:35
And I don't have the answer like how do we fix this?
21:38
But I think, you know, when you get people to start paying
21:41
attention to it, then there'll be some things that are put in
21:44
place to to maybe reduce the instance some.
21:47
Absolutely. I think you said it perfectly
21:49
earlier on in our episode where you said that these are the
21:51
silent killers. These are the things that we're
21:53
not necessarily noticing or we are kind of brushing off as one
21:58
off instances or cases, but they really do compound and set up
22:02
over time and create a multiple millions of dollars loss within
22:07
our industry, which is really significant.
22:10
When you think about our industry as a whole, especially
22:12
here in the United States and the economy that we're in with
22:15
all of our rancher populations and everything like that, that
22:18
adds up and it really does provide an impact.
22:21
As we kind of wrap this up, Chase, when you're looking
22:23
forward in the next 5 to 10 years, what do you kind of hope
22:27
that this technology leads towards?
22:29
What do you hope that you're kind of looking for in the
22:32
industry as far as just managing health risks, trace the
22:35
efficiency, any of the other topics we kind of talked about
22:37
today? Yeah.
22:38
So I mean, immediately, I think this this type of thing could be
22:42
used to help with building certain types of selection
22:46
tools, whether that's through a breed association or a genetic
22:49
test. Those things require a ton of
22:51
phenotypes and can take several years to build.
22:53
But with these types of tools, you can collect thousands of
22:56
phenotypes a day. So automatically that's that
23:00
part of the industry. Like I said, traceability, I
23:03
think, you know, it's common no matter what.
23:06
So I think if we have tools in place that people who care about
23:10
the industry and cattlemen build and that's that's really good.
23:13
So yeah, I know that topic scares people and I don't want
23:16
to touch on it too much, but I think it's important that we
23:19
start to kind of conform to some of those things a little bit.
23:22
But really, you know, these types of machine learning tools
23:27
and AI is perfect for maximizing efficiency.
23:31
And you know, the beef business, it is a pretty low margin
23:34
business. Depending on who you asked this
23:36
last year, maybe it didn't feel that way, but it'll come back
23:39
around. And when those margins start to
23:42
close, you have to be very careful about your inputs and
23:45
outputs. And these tools are perfect for
23:48
identifying weaknesses in identifying where you can
23:51
improve. And so I think just making these
23:54
tools more available a very user friendly, so people can, you
24:00
know, use their phone to have this type of capability or on
24:03
their own home computer to make better production decisions and
24:07
just boost their efficiency. Absolutely.
24:11
It makes me excited for where this industry is going and what
24:14
we can kind of see in the coming years.
24:15
Chase, as we wrap it up, do you have anything else that you
24:17
would like to say or leave with our audience members today?
24:21
I know, I mean, I appreciate you having me on today.
24:24
I love to talk about this topic. So you know, if anybody ever has
24:29
any questions, you can look me up on the university page.
24:32
I also have a sub stack page where I post kind of non
24:36
academic type articles. If you want to look me up on sub
24:38
stack, it's free. So I talked about congestive
24:41
heart failure and all types of other things so.
24:43
Wonderful, I will have Chase's sub stack and all of his other
24:46
links linked down below if you guys want to get in contact with
24:49
him and discuss any of his research or what he is kind of
24:52
doing or just follow along. Chase, thank you so much for
24:55
coming on and sharing your insights, your knowledge and
24:58
everything else that you are bringing to the industry.
25:00
To all of our listeners out there, thanks for tuning in and
25:03
we will see you in. See you on another episode of
25:05
the Cattle USA Daily Podcast next time.
25:07
Have a good one y'all.