[music]
Herbert Y. Kressel, MD Hi. This is Herb Kressel and welcome to the Radiology podcast.
Today I'm joined by Dr. Andrew Rosenkrantz, Associate Professor of Radiology and Urology
at the New York University School of Medicine.
Dr. Rosenkrantz and his colleagues authored a fairly definitive paper entitled "Diffusion-weighted
MR Imaging for Detection and Assessment of the Aggressiveness of Prostate Cancer: Comparison
between Conventional and Kurtosis Models."
Welcome Dr. Rosenkrantz.
Thanks so much for joining us.
Andrew Rosenkrantz, MD Thanks so much for having me.
Sure.
Many of our readers may only be peripherally aware of kurtosis, diffusion kurtosis imaging,
and its impact in assessing the aggressiveness of prostate cancer in particular, so could
you tell our listeners and viewers what actually is diffusion kurtosis?
Okay so I would compare diffusion kurtosis imaging to conventional diffusion-weighted
imaging which assumes that the water diffusion is free in its behavior.
Diffusion kurtosis imaging recognizes a broader range of hindrances to water diffusion and
can account for those hindrances in its modeling to extract information about the microstructural
complexity of tissue.
It can look at how complex the tissue is at the cellular and even the intracellular level.
So the more complexity of the tissue the less free the diffusion is and that would affect
the kurtosis.
It also affects the distribution of the diffusion parameters.
Correct.
Okay.
With that as a basis, why would you think this kind of a metric might be of value in
assessing the aggressiveness of prostate cancer in particular?
So for assessing the aggressiveness, prostate cancer we're really talking about the Gleason
score which currently it's a score that the pathologist determines on their visual
assessment looking at the tissue under the microscope.
That is really driven by the architecture, the complexity of the tissue in its appearance.
The conventional ADC is kind of more of this hodge- podge or amalgamation of facts also
influenced by tissue viscosity, other factors, and the hypothesis is that the diffusion kurtosis
is more of a kind of a pure or less contaminated reflection of really just the structural complexity
and hindrances which its believed could more reflect what pathologists are trying to estimate
when they assign the Gleason categories.
Thank you.
Now this study is a follow-up to your own which we published in 2012 was a preliminary
assessment of diffusion kurtosis imaging in prostate cancer, and since that one was published
there have been several others published and the results as you note have been mixed.
Why did you think that this current study you're reporting could more definitively
determine the utility of kurtosis in looking at the aggressiveness of prostate cancer?
Were there particular problems with the prior studies?
Yeah, so the studies I've been – so this has been a topic of a lot of interest in the
last few years, we're all trying to best predict the Gleason score non-invasively which
has huge implications for prognosis and treatment selection and the studies have had mixed results.
Some have been really promising, others have not been.
So in terms of some of the limitations of what's been done is the sample sizes have
generally been small.
There were some use of the reference standard biopsy instead of radical prostatectomy as
we did in our initial feasibility study of kurtosis in the prostate which has diffusing
biopsy has limitations in terms of the precision of the grading, and even just some technical
factors in terms of how the kurtosis image sets are processed and the metrics are obtained.
I think some of our understandings of this have evolved and improved over the last few
years, and I think in terms of what b values from a diffusion acquisition are incorporated
and how this is done.
It's not just pushing a button.
There's a lot of optimization of the (inaudible.)I think even that has gotten more precise as
well.
Good.
Can you briefly outline what you actually did in your study?
Tell us a little bit about the design and sort of what you looked at.
We had a sample, a retrospective sample, of several hundred patients.
At our institution our standard diffusion acquisition provides the necessary information
to be able to extract the kurtosis metrics.
All the patients underwent prostatectomy after MRI.
We included patients where pathology had done detailed mapping of the distribution of tumor
within the prostate including localization and grading of individual tumor foci.
We then selected the dominant or index tumor in each patient, and correlated these back
to the imaging and placed regions of interest to obtain what we deemed to be the optimal
ADC, standard ADC, and the optimal kurtosis metric that we could generate using our acquisitions.
Were these mean ADC and mean kurtosis over the whole volume?
It was the mean value of an ROI on the higher lesion.
And then we correlated that back to the pathology and we looked at the ability of ADC and kurtosis
to separate benign versus malignant as well as to separate low grade from high grade tumors
using two different definitions of high grade that have been applied in the literature.
So really we had three different histologic outcomes for which we compared conventional
ADC and kurtosis.
Okay and what were your key findings?
In terms of our findings, we found these interesting.
ADC and kurtosis they were highly correlated.
As ADC goes down, kurtosis goes up.
Both of those show as being generally associated with greater aggressiveness.
Their accuracy for predicting our various end points, benign versus malignant or high
grade tumor, were very similar.
When we looked at the subset of patients let's say there was a difference between them, kurtosis
actually did not outperform ADC within that subset.
So either they generally gave – for most patients they predicted the same thing.
Either ADC and a kurtosis said could be benign or be high grade or both of them said there'd
be, they both gave similar results or basically giving – to say it somebody's giving some
redundant or repetitive information from the kurtosis that we already have from ADC and
in the small minority when they were discrepant, the ADC actually did similar or even a little
better.
Although in the paper you describe what I would consider a fairly rigorous protocol
for matching the images particularly the target lesion with histopathology.
In 36 patients the target lesion couldn't actually be identified on the MR images.
Were you surprised by this and why do you think it might have happened?
So that's a really important point.
Those were overwhelmingly low grade tumors.
These are Gleason 6 lesions.
This is recognized actually in prostate cancer, this is a recurring issue whenever you try
to do detailed MRI pathology correlation for prostate cancer.
And other groups have described, other (inaudible) have described this as well.
Some prostate tumors they don't really form a discrete mass like say a renal cell carcinoma,
it's just it's histological sparse.
The malignant glands are interspersed with benign glands.
There's really no actual discrete tumor and these can actually just be invisible on
imaging.
For those we had no clear correlate on the imaging.
And for those in our final analysis, we ended up basically looking at the numbers both ways
where we just empirically placed an ROI and where is seemed the lesion was based on pathology
versus just excluding those.
Whether we did an empiric ROI or we left those out we had similar.
Our final observations held true.
Okay.
Now although the ADC and the kurtosis findings were correlated and concordant and around
80%, as you noted they were discordant at around 20% and you mention that in these the
majority of them the ADC was sort of more accurate or at least correlated better with
the histopathology, but my question is what do think is driving the discordance and what
sorts of lesions did you notice this discordance?
ADC and kurtosis they were correlated but these aren't – there are differences.
The ADC does have influences from viscosity concentration.
There are some other – it is kind of more this kind of this messier metric whereas the
kurtosis is really more tuned into complexity and hindrances to water motion.
There will be a small fraction where they didn't exactly follow each other.
I can't say exactly what the unifying kind of property was of the lesions where that
happened, but that it was not common and the ADC was either similar or a little better
than kurtosis in that.
So what you're saying is that they are in fact, although they're related, they're
measuring somewhat different things and it's not surprising that the measurements wouldn't
be the same in some subset of patients.
Exactly.
Now as we said in your study you looked at whole ROI or most of the lesion volume ADC
and kurtosis.
Others have started to look at subsets doing histograms of the ADC and the kurtosis looking
at sort of the worst actors, the bottom 10%, bottom 25% and have you looked at other kurtosis
based metrics?
So I think the histogram metrics for extracting a summary value from diffusion acquisitions
is really promising.
So say looking at the 10th percentile or (inaudible) this has shown I mean some really compelling
results in the prostate as well as in a range of other organs where work from other centers
or work that we've done at NYU.
I think unlike the diffusion kurtosis technique this doesn't really require any changes
or adjustment in the protocol or a standard protocol can be done and it's all on the
post processing side.
I think vendors are making those tools available and this is I think becoming easier to incorporate
into clinical workflows.
So I think this has a lot of potential in the near term.
The same thing with the kurtosis method or a histogram of it?
And this can be completely applied to the kurtosis data set which we didn't do in
this study, but I mean it's a great idea.
The literature has consistently shown incremental improvements in performance when applying
the histogram analysis.
So whether done to a conventional ADC, you could do a conventional diffusion analysis,
it could be applied in a comparable way to the kurtosis data set.
It wouldn't surprise me if that led as well to incremental improvement there.
In this study we were just doing a comparison of conventional mean between kurtosis and
standard diffusion, but I think I would anticipate that applying a histogram approach to either
of these could bump up their performance a bit.
Yeah I mean histologically as you know the Gleason grade can vary throughout a lesion
in scattered areas and so it would seem that sort of you know the and my understanding
is that the worst grade drives the final grade so that this might more closely mirror what
the pathologist is looking for and sort of finding the most aggressive areas to establish
the grade.
I agree with you, it sort of makes sense as an area for future investigation.
This sort of leads up to my final set of questions for you, it sounds like you haven't given
up on kurtosis yet.
Are you still acquiring this data in your patients?
We're looking at, I think more can be done to optimize it and there might be some additional
ways of, some additional angles that haven't really been as well explored or considered.
So no we haven't given up on it.
I think it can be something that can be combined with other innovations.
As an example, more recently what we're turning our attention to is time dependent
diffusion.
This is a parameter that hasn't gotten quite as much attention by radiology investigators.
A lot of the focus has been on the b value and directions of the underlying diffusion
acquisitions.
There's also the actual diffusion time.
The time between the diffusion (inaudible) gradients actually can be manipulated.
It's playing with the actual source sequence.
The vendors don't always make as readily available to us as radiologists to work with,
but it can be, it actually can be adjusted and we're coming back and actually – we
can actually push it to really short diffusion times, shorter than are usually applied and
can technically we tune the contrast of the diffusion image sets in that manner, we can
actually go back and combine that with the kurtosis analysis or the conventional diffusion
analysis; and then we're finding some really interesting results that way.
So we're still using kurtosis but the idea is to combine this with some other manipulations
of the diffusion acquisition and we'll see what it shows over time.
Sounds like there may be another Radiology podcast in your future as you get more into
this work.
Dr. Rosenkrantz it's been a pleasure chatting with you about this paper, and I think at
this point in time it certainly gives us a good handle on the relative role of diffusion
kurtosis imaging in looking at prostate cancer aggressiveness.
Thanks so much for joining us.
Thanks so much again and I really enjoyed participating.
Great, bye-bye.
Bye.
Không có nhận xét nào:
Đăng nhận xét