9 Diffusion Weighted Imaging (DWI)

Introduction

Our next speaker is Andy Rosen Krantz.

Andy is associate professor in the Department of Radiology at NYU Langone medical Center.

And he's gonna talk to us about diffusion weighted imaging.

I am Andrew Rosen Krantz from NYU, and I'll be speaking about diffusion weighted imaging in prostate MRI.

Clinical Examples

So to just begin with a couple of examples.

This is this was a biopsy naive patient with an elevated PSA and I start with our T two weighted image, and you can take a look and see what you might think.

Maybe there's a more suspicious area here on the right.

But as we move to the image, the image sets from diffusion weight imaging, the A DC map, and our high B images, we'll go over in detail in this talk.

The dominant parent focus is this rounded circum front area of low A DC, an increased signal on the high B image, which many of us likely would not have identified as a dominant lesion on T two weighted image alone and on effusion biopsy.

This was a recent in three plus four tumor.

And as for a transition zone example, this was a patient with a prior negative biopsy.

And we can look at the T two weighted image.

There's some diffuse kinda just mildly reduced T two signal throughout the peripheral zone.

Nothing really standing out the transition zone on the diffusion weighted image sets.

Again, rounded circuit area of low A DC, an increased signal on the high B image.

Perhaps there was some low T two signal in that region in correlation, but harder to identify that with the T two alone.

And on tardive biopsy, this as well is a Gleason three plus four tumor.

Challenges of T2-Weighted Imaging Alone

So I think this brings out the challenge of T TTU imaging alone, that there's often a heterogeneous appearance of the peripheral zone and TTU image with numerous benign causes of decreased T two signal inflammation, atrophy, scars, hemorrhage.

These are patients who have a heterogeneous appearance of the prostate on the T two image, and neither have been shown to ever have tumor on biopsies that they've undergone.

And this non specificity of T two imaging limited the utility of prostate MRI prior to the emergence of modern multiparametric imaging of the prostate.

Introduction to Diffusion-Weighted Imaging

And this leads directly into pros to, into diffusion weight imaging, which has substantially improved the ability of MRI to localize tumors in the prostate, and is a key region reason for the surge of interest in prostate MRI in recent years, that this is now considered as an essential component of routine prostate MRI protocols.

And there's various factors that play into this, and also radiologists are able to do this well.

I mean, it's a non-contrast technique.

It can be obtained relatively quickly compared to other functional parameters.

We can acquire and process it with modern MR equipment and without direct physicist support just very quickly around its background.

So this is intended to reflect tissue cellularity, which is increasing in the setting of neoplasia, just kinda a cartoon-like schematic showing relatively free diffusion in benign tissue compared with more restricted diffusion of water molecules in the setting of increased cellularity in a tumor.

Technical Acquisition of Diffusion-Weighted Imaging

And the way we do this in practice is we obtain diffusion weighted images at different B values.

The B values is the strength of the diffusion weighting, and the signal and tissues will vary across the B values and the scanner.

The systems can inline as the sequence is being run, will can do computations of the signal intensity versus the B value and give us the a d c or apparent diffusion coefficient, generally using a mono exponential fit.

And as we go to higher B values, we see wider separation between benign and tumor.

And this really reflects the it's essentially the inverse of the slope, and then the tumors end up having the lower a DC values.

So typically in practice is acquired using a single shot echo planar sequence.

Just looking across recent articles and literature and the methods and the in terms of the technique being described for diffusion, it was somewhat variable.

Many signal averages are needed to obtain sufficient SNR, particularly on the high B value images.

This is from what I was just kind of reviewing and putting this together was ranging from four to 12 sigma averages.

But I would suggest being towards a high end of this range, I mean eight to 12 sigma averages, I think it's worth it to invest the time to paint a high quality a DC map given the its central role in our assessment.

The total scan time again, it is variable I've seen in the in papers two to 10 minutes, but again, we're towards the upper end of that range, and I would again advise spending the time to get the highest quality diffusion weight images you can.

The slice thickness orientation typically will match T two weighted images, and a high inflated resolution can be helpful within the limits of that your SNR.

So again, here we have in the same patient kinda a low res and a higher spatial resolution diffusion weighted image.

And this was the patient's proven tumor and more apparent with the higher resolution minimum two B values are needed for a standard mono exponential fit.

Low B value can be done with zero.

But as mentioned in the prior talk going to slightly above zero, say 50 or a hundred can be helpful given perfusion contamination of b value of zero high B value typically have been around 800 to a thousand in much of the literature up until the last few years, as we'll talk about many centers use an additional mid B value, say around 500.

That's not essential.

And the modern scan current scanners will typically provide automatic reconstruction of the A DC map as the sequences are being run.

Assessment Using the ADC Map

And in terms of assessment, it really is the a DC map that that's that's the crux of how we interpret these cases.

And kind of in the late two th kind of 2007, 2008, I cite a couple papers, but additional papers around that time that we're all kind of making a similar observation and reporting similar findings of how of the extent to which we improve our accuracy in localizing tumor using the A DC map.

Here we have two patients.

On T two 80 image, there's decreased signal bilaterally.

Again, there could be tumor in either lobe in both lobes in neither lobe.

We look at the a DC maps on these patients in the top patient, again, the focal round of low A DC on the left and in the bottom patient on on the right.

And these were the patient's dominant tumors and kind of a string of papers around that time showing substantially improved accuracy using the A DC map.

Here I show the data from one of these where they had three independent readers, and for all readers, they had significant improvement using T two plus A DC compared with the T two weighted image alone.

Integration with High B-Value Images

So how does this integrate with our B value image?

This is one of the questions during the earlier break.

We do always look at the high B image, but at least if using B values in the range of say 800 to a thousand, you're gonna see more lesions on the a DC map than on the high B image.

Here's a dominant lesion on T two and on the a DC map, what happens on on the B value of say 800 to a thousand, the benign prostate often still is not fully suppressed and will have some mild increased signal.

So trying to differentiate the tumor from that might not always be so apparent.

This was kind of an older study.

This was at 1.5 without an endo rectal coil.

So the sensitivities aren't super high, but there was a higher detection of lesions on the A DC map than on the B 1000 image.

Ultra-High B-Value Images

I think in the last few years this is changing.

There's been now a string of studies.

I cite several of these, but there's been more going back to about 2010 using what I'm labeling here as ultra high B values, which I would consider to be B values in the range of 1400 to 2000.

There's greater suppression of the benign peripheral zone with B values in this range, which can improve our conspicuity for tumor.

So this patient had a proven significant tumor in the right transition zone B 1000.

The entire prostate is bright.

This lesion might not stand out to your eyes at 1500.

Much of that signal is now dropped out, and the lesion is has greater conspicuity.

So there's different ways of doing this.

The ultra high B value images might be challenging to directly acquire given low s and r, as well as increased distortion at the very high B values.

So this can be done with just basically extrapolating or computing them from the lower B values.

These are basically mathematically extrapolated from that curve we were showing earlier.

And this will give the image contrast of the ultra high B value images, but the quality in terms of distortion artifacts of the relatively lower B value regime.

And this doesn't require then additional image time or acquisition time compared with just acquiring the standard B values.

So here we have an acquired B 1000 image.

This patient had a proven dominant lesion in the left, or tumor in the tumor in the left posterior peripheral zone at B 1000.

The entire prostate is still bright.

While this area is relatively brighter at 50, an extrapolated B 1500, much of that signal in the benign prostate is now suppressed and the lesion has greater conspicuity.

This was one of the images from earlier I had shown the B 1500 image extrapolated in comparison back with the acquired B 1000.

The lesion doesn't stand out as much.

So this was the data from one of the studies.

They had three independent readers, and again, for all of the readers, their accuracy went up using computed B values of 2000 in comparison with B 1000 images.

So the utility of these I think is well established in these recent papers for when directly assessing the native B when directly assessing the B value images that our performance improves with these high B values.

It's less clear if the a DC map computed from the ultra high B value images will be of greater utility.

By using these higher B values for generating the a DC map we do need sufficient SNR at the maximum B value to reliably do the a DC computations.

And again, our SNR drops as say we go to a very high B value.

So one possibility is to essentially do two separate diffusion acquisitions in your protocol to do an initial acquisition with a maximum B value of about 1000 from which the A DC map is generated.

And then outside of that, again, this is if you're not gonna if you're not doing computed but looking to acquire it, then outside of that directly acquire the ultra high B value image, but not use it for the A DC computation.

I think these ultra high B value images can be a particular value in the transition zone where we have a lot of pitfalls from say, stromal BPH nodules.

These stromal BPH nodules can have a low A D C and can have some overlap with tumor.

So we're trying to distinguish say, a stromal BPH and a transition zone tumor looking for persistent increased signal of the very high B values can be helpful here in the A DC map.

So some stromal B patient nodules will be quite dark in a DC, looking at this was a B 1500 computed, and looking and seeing the persistent height signal that can raise a confidence in suggesting it as a candidate to say target rather than just say dismissing as BPH.

And then this was LEHAN six on a fusion biopsy.

This was a patient of prebio MRI T two, and a DC.

Perhaps nothing would stand out or nothing here that you may call on a B 1500 image.

There's this curva linear area of increased signal.

The anterior left trans anterior left lobe.

This underwent fusion biopsy was actually Gleason nine tumor.

It was actually quite scary because with the T two and a DC alone, perhaps would be hard pressed to really suggest that finding.

So this was the same paper from earlier, and they also broke down their findings by zone.

And in the transition zone as well, just in isolation.

All three readers again also had significantly improved accuracy using the ultra high B value images.

Application in PI-RADS Scoring

So in terms of bringing this into our assessment here, here, I'm this is merely from Pyres version one, but the point I'm making applies as well to version two.

The they were in terms of their criteria for the one through five assessments for diffusion is based on finding on both the a DC map and the high B value images that both image sets have to be looked at carefully.

And whether something is one category or another is influenced by the combination of findings across the two image sets.

So in practice this is a patient.

So in terms of how these scores are useful, this was a patient on active surveillance.

Recent six baseline study we had we described here some we call this a pires two, kind of a some geographic finding some mild decreased signal, not circumscribed on a follow up exam.

Two years later, the finding was larger and had more pronounced or clear cut a DC reduction.

We upgraded this to a PY RADS four.

It was more of a focal lesion.

And interestingly during the time the PSA it was the only current really the only marker in clinical practice for monitoring is patients noninvasively was stable.

Quantitative ADC Values as a Biomarker

So I think if we can move beyond just a kind of a qualitative assessment of diffusion weight images and also look into how to best apply this quantitatively, there's been a lot of investigation in the role of a DC values as a biomarker of tumor aggressiveness, and the literature has actually been really favorable for a DC as a biomarker.

This is based on what's been demonstrated and demonstrated inverse correlation between cellular density and a DC that as the cell counts within tumors go up, the a DC value has decreased.

In this study, the correlation is quite good.

So there again, this is another area where there's been a string of papers, one after another kind of validated validating this association.

This is just one, but there's many similar diagrams and figures that I could show.

Here we see progressive stepwise decreases in a DC values as going from benign prostate to low intermediate and high grade tumors.

This was a very interesting study.

They compared the ability they looked at the ability to predict the overall the final let score and radical prostatectomy between just standard trust biopsy, which here performed poorly, had an area under the curve of about 50% or of a coin flip versus non-invasively trying to predict presence of high grade tumor via just the a DC value in the dominant lesion having an a UC of around 80% so substantially outperforming biopsy in predicting the final Gleason score.

Here was this was using a DC values to predict findings on targeted biopsy, and again, showing a similar associations.

This was for separating progressors versus non progressors as described in this paper for patients in active surveillance.

They they did some more complex modeling with vast and slow a DC components.

But regardless of the approach used, again, there were differences in the baseline a DC values amongst those who did or did not quote progress in that paper.

This was a very interesting one.

Looked at predictors of biochemical recurrence after radical prostatectomy here ranked in order of decreasing significance, at least at Univar assessment.

This has all the standard feature stage a tumor size, volume, Gleason, PSA, and in nursery, the best predictor was the tumor a DC.

One issue that comes up is, so if we're trying to apply these numbers in practice, actually measuring a DC values as we read a case to integrate that into our reading, is there actual threshold or a cutoff that comes up that comes up and different people different papers out there are suggesting different numbers, but they're kind of clustering a similar area of around 0.8 to 1.0.

In the units listed here.

So this was for instance, a study actually from the last speaker Dan Margolis.

And they they proposed a threshold around 0.85 and showed improved performance for separating significant versus insignificant cancer in comparison with just traditional Epstein's criteria alone with without consideration of MR findings.

So you know, if looking to report this, for instance, this was a patient on surveil on set Gleason six tumor on MRI, we call it a RADS four.

So I guess we could say this is under 1.5 centimeters in the right peripheral zone.

It's a DC though was 0.72.

So we could apply that threshold, for instance, and suggest it's a high grade lesion.

And on targeted biopsy, this came back as Gleason four plus three in comparison.

This also was a surveillance patient, Gleason six again on MRI, we described a PYS two in the left.

This wedge shaped area of geographic mild decreased T two and a DC.

So it's not on the slide but th this had it's a DC value oh, I'm sorry, here.

This A DC was 1.3.

So that would be above the threshold.

So just again, benign or a low grade lesion.

And at least in our practice, we would just continue active surveillance with monitoring A PSA and MRI in a year.

And no repeat biopsy at this time, given the MRI findings.

So whether this youth a DC value is ready for routine use, I mean, there's still issues.

There's still some problems with this.

There's while it's great and when it looks very good when comparing groups or populations, there are still a lot of overlaps.

It's the same diagram from before.

And we can see if you look at these error bars are quite wide and going all the way from the benign through the high grade lesions.

There was a large range of overlap that this a lot of this comes from significant inner patient variation in the a DC values of the benign peripheral zone patients will just vary in how bright or dark their peripheral zone will look.

On the a DC map, this is a patient who's had negative biopsies and just has a diffuse reduction in his A DC in a in the peripheral a DC throughout.

Couple of interesting papers recently that sort of improved accuracy for identifying high grade tumor by normalizing the A DC relative to the a DC value in that patient's benign appearing peripheral zone and looking at ratios between a d c and the peripheral zone and showing improved accuracy with this approach.

There also have been suggested maybe apply more advanced models than just the more of a the crude mono exponential fit descriptions of g non Gaussian models.

The IVM there there's various models being applied.

This approach is supposed to look more complex tissue microstructure and showed some improvement compared with standard A DC.

I think there's a lot of interest in these, but still by and large in current clinical practice.

And what the scanners are offering us is still a a DC today in terms of optimizing the images.

Optimizing Image Quality and Artifacts

So obtaining high quality diffusion weight images in the a DC map is critical for reliable and accurate interpretations.

And this is the sequence most prone to distortions and warping with rectal gas that's been mentioned being the key contributor.

And these distortions are possible even with a relatively empty rectum.

I mean this is what we might see where there's some kind of this non anatomic warping of the prostate anatomy.

So the rectal gas is gonna be a source of potentially severe artifacts that can render diffusion.

Non-diagnostic and various patient preps have been attempted to achieve an empty rectum and reduce bowel motion.

These approaches are inconsistent between centers.

We all kind of have our own favorite technique or what's worked well for us or what we like to do in our practice.

There isn't really much data regarding the optimal technique.

So I can kind of just describe here what some different options are out there, and you can see what works well for you in your practice.

But I think we said there's really no clear solution even with any of these ones that I'm describing.

I don't know if any of these will necessarily completely resolve this problem.

So so just to see like more of the extreme, because here we have a very distended rectum.

And if we look at the images, here are the prostates essentially almost completely efface on the axial hy be image and a DC map kind of squished.

So things that we can try doing an enema or a laxative even for a non endorectal examination non endorectal coil examination.

A dietary approach is advising the patient to say have a bland diet or avoid hot caffeine beverages have been suggested suctioning the er recal catheter prior to the examination instruction to evacuate completely before the exam.

Maybe glucagon for battle peristalsis.

Again, various things, but I would feel hard pressed to say any of these will completely eliminate the problem.

Also in terms of the scan parameters, even just beyond the prep these parameters can be adjusted to minimize susceptibility artifact and distortions.

So using parallel imaging, minimal te possible increasing the receiver bandwidth left the right face encoding.

So to some extent, a none of these are gonna be routine and might be the default settings, but you can at least check or confirm that these are applied in terms of the face coding direction at least for a non indirect coil exam here.

If we look at the size and of the and shape of the pro the T two weighted image we have a a DC map with a to P phase encoding, and it's somewhat compressed or a faced relative to the T two weighted image if switching it to left to right phase encoding.

Now that that's improved, we kind of it's fuller here.

It more closely approximates or matches the dimensions of the prostate on the T two weighted image.

Pitfalls in Interpretation

In terms of interpretation.

Now, just trying to go over some briefly, just some kind of pitfalls.

And some of these have been mentioned in other talks.

So prostatitis or inflammation can be a cause of false positives, not just on T two or DC as well can cause reductions in a DC.

Here we see a focal reduction in signal on the T two as well as a correlate on the a DC map.

This underwent fusion biopsy and came back as focal in inflammation on a follow of MRIA year later.

There's just there's some diffuse geographic decreased T two signal, but the focal lesion is no longer appreciated.

Sometimes you just can't so some cases they're gonna be very hard to tell the difference reliably and can really can just say there's a suspicious abnormality and without biopsy won't be able to be sure.

In the transition zone, again, this has been mentioned that b uh bph H or SHRM BPH nodules can have low ADCs overlapping with tumor here.

This nodule had low AD C and actually did actually end up getting biopsied and and was just and was BPH h So again, as been mentioned, T texture morphology and T two weight image is most helpful.

If I'm using ultra high B values can be helpful for separating stromal bph h and uh transition zone tumors.

And if are if you are using a DC, then might need a different threshold.

So just run through some additional pitfalls.

So here on the a DC map 'cause these are of rounded area of decreased A DC here on the posterior left peripheral zone you know, could this be a suspicious lesion?

If you when you look back at the other sequences, this is actually outside the prostate.

This is part of the neurovascular bundle that sometimes an a DC map it can it's a little distorted.

It can be difficult to appreciate.

It might actually look like it's inside the prostate if it's really sitting right along the capsule.

So having to look very carefully for could some neural structure or vessel close to the prostate be a mimic.

Sometimes right by the boundary of the peripheral and transition zone, there's this kind of a loose fiber pseudo capsule that can be a little asymmetric, have areas of thickening that can create some fake outs right by the border of the two.

And this can be suggested to be a lesion.

Also just getting correct the windowing of VADC map, I think having tight windows and really bringing out the contrast to so you can really appreciate these lesions.

So sometimes the default windows that might come through from certain vendors or on your PAC system might not be optimal.

Might have to really find that the window level settings are optimal for your system to appreciate these lesions.

Role in Staging

So finally just a few points about the role.

There's gonna be a talk tomorrow in staging specifically, but there is literature showing the role of diffusion from proving tumor staging.

This study showed improved staging with diffusion compared with T two alone.

Again, show showing improvements in numerous measures of performance in that there was a lower a DC in tumors with extra pathetic extension in comparison to those without in this study it it looking for sensitivity for extra prothetic extension separating for focal versus more extensive or established extra prothetic extension.

And when incorporating a DC, the sensitivity for EC measuring at least two millimeters was a hundred percent for both readers.

And again, lower a DC in tumors with ECE.

And finally also been in a number of papers recently showing improved accuracy for semi vesco invasion when combining T two weighted images with the a DC map.

And in this patient here there was invasion of the base of the right seminal vesicle sh showing decreased A DC.

Conclusion

So in conclusion, diffusion weighted images substantially improve the accuracy of tumor localization.

Attention to the acquisition technique is important for optimal image quality.

The A DC map is the primary image set for peripheral zone tumor localization, while the ultra high B value images improve the conspicuity for tumors on the B value images and be aware of pitfalls in diffusion weighted image interpretation.

And I thank you for your time.