0:01

Artificial intelligence is something

0:02

that's really important to consider.

0:04

Applying to your chest cts, both

0:06

for lung nodule detection on most of your chest cts,

0:10

aortic cts and cardiac cts,

0:12

and particularly in the setting of lung cancer screening.

0:16

This review from November, 2022 looked at the available

0:20

artificial intelligence tools

0:21

for lung nodules on a special article in a series

0:24

on AI applications.

0:25

And as you can see, at that time there were a number of them

0:28

and there continued to be a growth in the marketplace.

0:31

On the number of these tools that are out there,

0:34

they can be used to detect, characterize

0:36

and track nodules over time.

0:38

They segment nodules and measure nodule volume.

0:42

And none of 'em, I would say is absolutely perfect

0:45

because this is a marketplace that is rapidly evolving.

0:49

And so this article did a really nice example of how you can

0:53

evaluate an AI tool and what it's good at

0:55

and maybe where it needs improvement.

0:57

So some tools, for example, not

0:59

so great at looking at ground glass nodules,

1:02

some tools measuring into the only not A 3D,

1:05

some tools incorporating risk calculators and

1:08

and using lung rads as built in.

1:11

So each vary in some of the details behind them,

1:13

but I think you get the gist.

1:15

It's detecting, characterizing,

1:17

measuring, and looking for growth.

1:19

They can assist you as the radiologist in the prognosis

1:22

prognostication for malignancy

1:24

and for categorizing lesions in the lung ran schema to date,

1:29

there actually haven't been large numbers

1:32

of scientifically published studies

1:35

of using these nodule tools in the clinical environment.

1:37

But we get a lot of feedback from radiologists about which

1:40

tools they like and which tools don't work

1:42

so well once they're implemented in their practice.

1:45

We have a, one of these tools implemented in our practice,

1:48

uh, from a vendor that we've been using for about a decade

1:52

and we've watched the tool improve with each iteration

1:56

of updates in the tool, which have gotten

1:58

to more consistent volume measurements,

2:00

more consistent nodule contours

2:02

and more consistent nodule measurements.

2:05

In addition to doing these tasks for you, you may want

2:08

to use some of the information that comes out of these tools

2:10

to put them into your radiology reports directly.

2:14

And we know there are lots of barriers

2:16

to adoption of these tools.

2:18

From AI being a black box, I don't understand

2:21

how it's making these calculations

2:23

to questions about the fear of the radiologist replacement.

2:26

I have very little concern about that one.

2:29

I think this is one more tool in our toolkit as radiologists

2:32

and we evolve in practice concerns about lack

2:35

of clinical validation, regulatory approval, reimbursement,

2:39

a tool that you'll pay for

2:41

that may not have any direct reimbursement line associated

2:43

with it, concerns about

2:45

how well it's integrable into your practice.

2:47

And these are all things that we need to understand better

2:50

as we apply tools like this in our practice.

2:54

Here's an example of an AI tool in practice.

2:56

This is one that we use.

2:58

It gives us a quick field report out of

3:02

where the nodules are, their size, their lobe,

3:04

and change by looking at prior exams,

3:07

whether they're our prior exams

3:09

or any outside chest cts

3:11

that we may have ingested into our PAC system for a patient

3:13

or it gives us a readout, but there are no nodules.

3:16

So this is comes with each chest CT

3:19

that we use in our practice,

3:20

whether it's screening or otherwise.

3:23

This is an example where there is a nodule

3:26

and when you drill down, it gives you an image of

3:29

where the nodule is in the greater image.

3:31

It gives you this output in a stack that you can scroll

3:34

through and link with your main axial images

3:37

that you're interpreting and it shows you the outline

3:40

that it has drawn, which allows you

3:42

as the user some comfort in knowing do you think it actually

3:45

represents the nodule as opposed to something that's kind

3:48

of in a black box and spitting out a measurement.

3:50

And then we have it set up to give us the average diameter,

3:54

the minimum and maximum diameter,

3:56

and it will also give us the nodule volume.

3:58

So these are the way that tools can be used in your

4:01

practice, finding nodules, measuring nodules,

4:04

and it will also report out whether it's solid parts,

4:07

solid or ground glass.

4:09

So here's an example

4:11

of an integration into our PowerScribe reporting system

4:14

where we get an example like this

4:16

where the patient has four nodules, you can see on each one

4:20

where there was no match to a prior exam

4:22

and these other three nodules,

4:24

the nodule was found on the prior.

4:26

You'll note comments like this may be on

4:28

the tools that you use.

4:30

User is responsible

4:31

for confirming nodule matches, for example.

4:34

So there is, it is incumbent upon the user

4:36

to validate the output of these tools

4:39

and by taking the pieces that come from the tool itself,

4:43

they actually automatically insert into

4:45

our power scribe report.

4:46

In that format. We talked about instructor reporting,

4:49

so it automatically puts it into our report, low density

4:54

size, image number,

4:55

and comparison size if it was there before.

4:58

The kind of things that are important when

5:00

we're looking at nodule.

5:01

So this is not an easy workflow to develop,

5:04

but when you do develop it,

5:05

it's very helpful in your interpretation practice.

5:09

Thank you for learning together about lung nodule morphology

5:13

and growth assessment and some of the tools

5:15

that you can use in your practice.