10 Dynamic Contrast Enhanced (DCE) MRI

Introduction

My name is Pete Choki,
and I'm going to be discussing dynamic
contrast enhanced MRI.

Outline of the Talk

The outline in my talk is that we're gonna talk about
what is D-C-E-M-R-I, for those who aren't familiar with it,
what does it show?
How is it analyzed using subjective,
something we call curology and modeling?
And then how useful it is in various situations?

What is Dynamic Contrast Enhanced MRI?

Dynamic enhanced MRI is the Rapid 3D acquisition of T one weighted scans, gradient echo scans before, during,
and after the administration of a bolus of contrast.

What that means operationally for prostate cancer is
a series of three to five baseline scans so
that you get an adequate assessment of baseline
before contrast.
Then acquisitions at about five
to seven seconds, no more than that per acquisition,
encompassing the whole prostate.
And then a bolus administration
of at least three ccs per second of a gadolinium chelate,
and then imaging for a period of time, usually
on the order of five minutes.

I don't expect you to copy this, but
it is in the printout, the syllabus,
but the key issues are that it's a 3D gradient echo image.
We typically use a fairly thin slice,
and most importantly,
the temporal resolution in our scan is about five
seconds or so.
With that, you can get very rapid lot
of images in a short amount of time.

Theory Behind DCE-MRI

The theory behind this is
that the low molecular weight contrast agents
that are very small molecules escape from the permeable
vessels associated with cancers.
That's true for gadolinium chelates as well
as Id donated contrast agents.

If you look at what happens within a lesion,
a typical cancer as a function of time,
there's a baseline acquisition where the scan
doesn't enhance.
Of course, there's no contrast.
And then there's a rapid period of enhancement followed by
some period of either washout or stability.
And of course, there are various ways to look at that.
You can look at the percent enhancement,
you can look at the time it takes for the maximum signal
to occur.
You can look at the percentage of washout from this peak
to some time point later on.
And then another approach is
to do pharmacokinetic quantification.
We'll talk about each of these.

Vascular Differences in Tumors vs. Normal Tissue

The usual normal vasculature of normal organs,
including the prostate, is one of very stereotypical anatomy
where arteries lead to arterials, to capillaries, to venial,
and then to veins in a very orderly process.
And the blood flow is efficient and rapid in tumors.
The blood vessels are chaotic
and often flow in both directions.
And their hallmark is that they're very permeable.
They have actual holes in the vessels that allow the egress
of small molecules very readily.
That's tumor angiogenesis.

The hallmark of cancer
with D-C-E-M-R-I is a rapid rise in signal, followed
by a relatively rapid washout,
whereas normal tissue has a slow rise in a signal
and tends to plateau or gradually accumulate.

Limitations in the Prostate

Having said that, in the case of the prostate,
there are some problems.
BPH is actually a highly angiogenic process.
It's still benign, obviously,
but it's driven probably by hypoxia.
And so there's a lot of new vessel formation,
and you'll see on almost every BPH scan, a lot
of vascularity within the BPH,
making it relatively less useful in the transition zone.
Prostatitis another common illness
that confounds MRI is also characterized
by highly permeable vessels.
You won't get any help there.
And so D-C-M-R-I has its limitations.
It's non-specific for cancer.

PI-RADS Criteria for DCE-MRI

In PI rads, the visual criteria for prostate cancer, as Katya nicely showed, was
that DCE is negative if there's no or minimal enhancement.
And also if there's diffuse
or nonfocal enhancements as would occur in prostatitis,
or it's clearly identified with a BPH nodule.
It's very well circumscribed nodule.
It's positive only when the enhancement occurs early.
That is typically within eight seconds of the arrival
of the bolus in the femoral arteries.
It's focal that is, it corresponds to a lesion on T two
or DWI, and it's clearly not associated with a BPH nodule.

As you have seen with the PI RADS criteria,
this really only affects a minimum number of conditions.
Basically, when the T two
and DWI are indeterminate in their respective areas,
the D-C-E-M-R-I can be helpful in pushing you towards
or away from biopsy of those lesions.

Clinical Examples of DCE-MRI Utility

You can see very small lesions with D-C-M-R-I,
this very tiny enhancing lesion.
It is seen, the hallmark on this CA loop is
that it enhances very quickly, very fo does correspond
to a lesion here on the T two.
And so depending on how the DWI looked in this case,
you might add to the risk based on this enhancement.

In typically, however, in most cases, the T two weighted scan,
in this case showing an anterior lesion
and the diffusion weighted scan confirming that,
the DCE, which is also very enhancing,
doesn't really do much to convince you
that this is a bad lesion.
Probably a PI RADS five.
So in that case, in many respects,
the DC is superfluous,
but as mentioned, sometimes these images aren't
so clear, and in that case, it can be very helpful.

Here are just some examples where I think
what you see helps you make a diagnosis,
but in fact doesn't push you beyond
what you already knew from the existing non enhanced images.

A very large peripheral lesion on T two with evidence
of extra capsular extension with that interruption
of the capsule.
And on the diffusion weighted image,
you can see basically the same thing with the extension,
very nicely depicted on the D-C-E-M-R-I,
this taken from the peak enhancement.
You actually lose information in this
because you can't see that extra capsular component.
And it really is kind of redundant
with what you already know.
Moreover, if you were just to look at the D-C-A-M-R-I,
you might think that this lesion actually extends all the
way across the back of the prostate, whereas it's really
more focal than that.

As I mentioned, there are confounders, in this case,
prostatitis.
When we give contrast in that case,
there is some apparently focal enhancement.
There is diffuse enhancement as well.
And eventually, it's very confusing to say whether there's
this is adding or subtracting.
You can see that there's sort of patchy appearance
to the T two weighted scan
and at biopsy, this was prostatitis.
So in that case, I don't think the addition
of DC really made a difference in moving
it one way or another.

Subjective Analysis: Curve Types

Many people have said, you're just looking at it in a very subjective way,
whether it's rapidly enhancing, whether it's diffuse,
what about trying to put more order onto what you're seeing?
And the simplest way to do that is
to simply look at the curves and what their directions are.
And of course, this is very founded in breast MRI,
where we have a type one curve
that shows persistently increasing signal,
a plateauing curve, that's two, and a washing out curve.
That's type three. And you can do that
with very simple software that allows you
to take a look at the images.

As you get more complicated,
you can start to compare that to the actual input function.
You can put a cursor on the femoral artery
and see that it's a very good bolus,
that there's a rapid increase in signal followed
by an expected washout.
And then you can look at the lesion.

One of the problems with this though,
and it was incorporated in the first version of PY rads, is
that it turns out there's a lot
of heterogeneity within any given prostate lesion, so
that you may be able to find curves that look like this,
that would be highly indicative of cancer
and curves that look like this, which are not
so indicative.
You could say, we'll pick the worst one,
but that may be only one or two pixels,
and would you make a dominant decision based on that?
In fact, the literature does not really support
this kind of approach.

There is addition software that
will support this.
Software that allows you
to put a cursor over a particular region
and get these nice curves.
As I said, depending on where you put the cursor,
you'll get different things.
And so the manufacturers have said,
perhaps we need something a little bit more
specific than these curve types.
And that is these kran
and KEP maps that you've probably seen.
And these are
so-called pharmacokinetic parameters.

Pharmacokinetic Modeling

What is a pharmacokinetic parameter?
It really dates way back into the 1940s when people
were trying to do tracer dynamics,
and they simplified the body into these compartments.
The one compartment is the blood,
where the contrast arrives into the blood,
and then it exchanges
with the extra vascular space.
And the rate at which
that happens can be parameterized in the forward direction.
Some people call it KPE,
but another more common descriptor is K trans.
And then as the contrast accumulates in the extracellular
space, it will start
to migrate back into the vascular space.
So there's a washout, and that is characterized by KEP.
That's very nice.
That's a simple two compartment model.
Many people believe that
that definitely oversimplifies what's going on in the body.
There are more than two compartments for sure.
But as a first approximation, this isn't that bad.
And then you can use color coding
to sort of get at that.

This takes us back to this curve.
How in fact do we derive these K trans and keps?
It's by actually fitting those parameters to this curve.
I don't want to first of all,
I don't really understand it,
but second of all, I don't want to take a lot of time by
trying to explain it.
But there are people who do.
And so you can have these equations
that are built in to the software
that you see.
And the key components of that are that you do need an input function, which is this CAP,
which is basically the input function.
You're gonna have to measure that off the femoral artery
or something like that.
And then you measure the ct,
which is the concentration in the tumor over time.
And then you fit these various parameters, kran
and KEP, in those curves.
And then there are actually three parameter models
that you can use that also require the input function and
will derive these kran
and KEP values.

A lot of work has been done on that.
What's what comes out of it?
For one thing, you can get very colorful maps.
Here's the T two and the a DC.
These are reversed and the DC that I showed you before.
And here you can see the very nice ca trans map
that really doesn't add significant information
to this already characterized tumor.
It does also show some incidental BPH that's
not very useful
and shows you some scatter over here corresponding
to an area that did not have cancer.

One thing
that we learned immediately from doing these tran
experiments with lots of people is
that there's a tremendous amount of heterogeneity
between patients in terms of the crans values.
Comparing crans from patient one to crans of patient two
is very hazardous.
The only thing that you can do
with these maps is really compare it to the patient itself.
You can look, you can see differences within the
background crans,
and those become apparent on these kinds of scans.
And of course, the parametric map is a nice thing.
It does enable you to pick up a lesion in this case
that might be otherwise difficult to see even
with your eye.
But as many groups have shown, it's very difficult
to correlate the exact tran's value with malignancy per se.
And so while it's a kind of quantitative technique,
it really doesn't help you in deciding whether
something is a cancer or not.
It has become less important over time.

This slide basically reiterates this point
that you can look at the raw image,
and this is just a screen capture from the peak enhancement
and see the area of enhancement.
Or you can do these maps and pretty much see the same thing.
Granted, they show you slightly different parts
of the curve, but in fact there's not a lot
of added value over the visual descriptor
or the visual impression
that you get simply from the cine ene loop.

Current State and Utility of DCE-MRI

This is the current state of affairs.
And while occasionally the DCE can really save the day, like in this case
where there was diffuse low signal throughout the entire T two,
the A DC was not that helpful.
But the KT trans map was,
or KEP map was able to really pull out this lesion
that we did end up biopsying.
And it was positive,
but these are really exceptional cases, maybe five
to 10% of cases will it improve the diagnosis.

This is a little bit of a busy slide,
and I apologize to it,
but it represents really the recent literature on DC MRI.
It compares those studies that just did T two
and DWI versus T 2D, WI
and DCE to look at the sensitivity.
And I just want to, these are group numbers.
In this case, 45% sensitivity without DCE,
49% with 76 to 81.
That's a five point spread.
Some studies had a greater difference.
But you can see it's on the order of five
to 15% differences.
And in many cases, there was no difference between the
DCE and the non DCE studies.

This has prompted PY RADS to really downplay
D-C-E-M-R-I.
While it's an endangered species,
it's not actually extinct,
but it's useful in a minority of patients.
Nonetheless, I think most of the people on the
steering committee for PI RADS still employ D-C-E-M-R.
I use it when the other techniques are disabled
for some reason, use it, always look at it just
to make sure that there's not some
highly enhancing lesion
that was just not perceived on the other scans.
It's a backup, a confidence builder,
and occasionally you'll see significant advantages to using it.

Future Directions: Radiomics and CAD Systems

I just want
to talk a little bit about some future looking aspects of D-C-E-M-R-I.
We've been through a few things.
We've been through Curology, which
didn't really pan out.
It's a very simple approach,
and we've been through kran,
which is a very complex approach,
but makes a lot of assumptions about the way the world is
and the way the human body is,
which are probably not accurate as we've discussed.
And yet we see, even with just visual detection,
that we can pull out some information.
There's an increasing school of thought
that there's a lot of information on these scans
that we're leaving on the table when we do strictly
visual analysis of them.
That's creating this industry, if you will,
of radios.
You might have heard that term,
or we call it decision support systems
in which you take the images,
T two A DCK, trans,
or even the raw DCE, you register them
to pathology to each other and then to pathology.
And then you can begin to pull out,
have the computer pull out features that are useful,
that associate the scan type
or the scan features with pathology.
And you do this in an automatic way.
And you can get things like texture analysis
that are simply not visible to human practitioners,
but are kind of good for number based
algorithms.
And these decision support systems are
coming in virtually every disease for a lot
of things, and they remain to be tested,
but they seem very promising.

We've been playing around with a few of them.
And this one in particular, of course this is a fairly obvious lesion on T two
and a DC, and the D-C-M-R-I,
but this is the cad decision map.
This is exactly what you would get at the end of the analysis.
You would see the scan that shows an increased uptake and it incorporates all this data.
None of it is really thrown away.

Here are three examples
of a CAD system pulling out a small tumor in the anterior part of the gland,
a large tumor that's fairly obvious.
And here's a smaller tumor
in the anterior part of the gland.
I believe this is sort of on par
with autopilot for jets.
You're still gonna be flying the jet,
but you use available tools to try
to draw out additional information.
And certainly DC MRI, not the way we interpret it today,
but extracting data such as what does one pixel next
to the other, how different are they?
These kind of things will enable us
to perhaps get more information
and improve the diagnostic capabilities.

I think another thing is that it will really help
inexperienced readers to get sort of into the game in
that they can probably perform at a moderate
or even expert level with the help of a CAD system.
I think this may be coming down the pike.

DCE-MRI in Recurrent Disease

I wanted to spend a few minutes on a play on something
that I think D-C-E-M-R-I is really useful in,
and that is for recurrent disease.
Recurrent disease is a problem
because the anatomy is disrupted,
and often the patient is on medications
and T two and diffusion.
Often there's not enough tissue there
to really make a judgment about whether something is malignant or not.

We talk about the stage
of prostate cancer called biochemical recurrence.
And this is either
after a prostatectomy where you see a rise in the PSA
of greater than 0.2, or
after radiation therapy in which the Phoenix criteria is a
rise of greater than two over the Nader that the patient
was able to achieve after radiation.

Both of these set off alarm bells
that the cancer is coming back,
so-called biochemical recurrence.

There's clear evidence that the earlier
that therapy is given to patients in the setting
of biochemical recurrence, the better they do.
And that is unfortunate for a lot of
the current pet.
It's like FDG
or fluoro choline, F-A-C-B-C, they're really insensitive at
that low PSA value.
The PSAs really need to get above one
or two for those scans to turn positive.
We're not gonna get any help in this setting,
at least from these kinds of PET scans.
There's evidence that other kinds
of more sensitive PET scans may be able
to play in this low PSA regime.

Just to show you at least one piece of evidence
that the outcomes in these patients depends on their pretreatment PSA.
If it's below 0.5 nanograms per milliliter patient,
these patients do relatively well below 0.5,
they do poorly, and the earlier you intervene, the better.

With that in mind, DC MRI seems to be able
to play a very important role.
For instance, this is a 45-year-old man with a high
PSA status post proton beam treatment
for Gleason three plus four disease.
And he has a strong hip family history.
And so there is an area of abnormality in the prostate,
but also in the seminal vesicle.
You can see the lesion in the prostate on a DC.
And there's increased signal in this area, in the prostate
and on the crans map in the sv.

This was all biopsied.
This is post-treatment of course,
so this is a recurrent disease,
and it was very high grade cancer at recurrence.

It's possible to do these.
This is a very interesting case of a patient
who received brachytherapy prior,
and had a rising PSA.
You can see not much is very difficult,
even though the asterisk is actually on the lesion on
this T two, very difficult to see.
And the a DC map is not particularly interesting,
but the DCE clearly shows an area of enhancement,
notwithstanding all the artifacts from the little seeds.

This was biopsied
and proved to be recurrent disease.

Another form of treatment that we're doing more frequently is focal laser ablation.
And in this case,
this was a low grade tumor anteriorly in the prostate,
you can see all three modalities show something.
And then under MRI, we put a laser fiber into the lesion,
and it's treated using Mr.
Thermometry. So how do things turn out
well on post-treatment?
T one, T two, you can see this kind of hole there.
The a DC map, not to informative,
but very importantly,
the post-treatment D-C-E-M-R-I shows very
little, if any enhancement.
There's little granulation tissue initially within the first
few weeks after this, but
after that, at six months like this, there's
very little enhancement.

Summary

To summarize, I think D-C-E-M-R-I plays a role in multiparametric, MRI,
but it plays a minor role,
and that's based on a lot of evidence that shows,
if anything, a small benefit to the addition
of DC in comparison with T two
and diffusion weighted imaging.

It is most useful,
and that's incorporated into PY RADS in equivocal cases.
So PY RADS three cases
where you're sitting on the fence when standard sequences
are, or when standard sequences are technically suboptimal.
So that becomes a useful role for it,
but kind of a backup role.

And in terms of these other ways
of analyzing D-C-E-M-R-I, either with Curology
or modeling, it seems that visual assessment does just
as well as any of those,
and that we look for focal early enhancement and it,
however, and I think this will be an important role in the future, it does appear to be very useful
for the detection of early recurrence,
after definitive treatment with, for instance,
surgery or radiation.

With that, thank you for your attention.