Quantitative Measurement: Techniques and Challenges - HD

Introduction

Thank you Professor Du.

It's the last talk before tea.

Your course organizer has, for better or worse, seen fit to let a physicist loose on you.

There's not too much physics, but my reason for being here to talk to you today is to talk about some aspects of quantitative measurement when you're doing contrast imaging.

Importance of Quantitative Measurement

Quantitative measurement isn't something that you've heard talked about, particularly in the guidelines.

It's something that's more of a research tool, and it's a really important aspect of the research that's associated with ultrasound contrast agents.

A lot of the techniques you've heard talked about so far today, you've seen pretty pictures.

You've seen the power of the images.

You've seen how these experts can show you details on these images and convince you that you can differentially diagnose different conditions.

But a lot of their expertise is built on earlier quantitative studies that allow people to analyze many cases and show that you had the evidence to back up these statements that now you can perhaps make more freely in a more qualitative way from an image.

That's one way that quantitative research is very important.

When you wanna prove anything new, when you all start to use contrast in your daily practice, you might come up with new ideas for how to use it, and then you might think, we need some way to quantitatively demonstrate these new ideas.

My reason for talking to you over the next few minutes is to just discuss with you or give you the opportunity to start a discussion about some of the factors that you need to have in your minds when you're doing quantitative analysis of contrast imaging.

Microbubbles as Blood Pool Agents

The most important thing and the reason why microbubbles lend themselves to this type of research is that they're blood pool agents.

We've heard lots of different talks today reminding us that under normal circumstances, the microbubbles don't leak out of the blood vessels or the cavities we put them into.

What does that mean? That gives us a tool for measuring things to do with the blood if we put them in the blood.

It means we can measure the volume of blood vessels in the tissue.

We can measure how fast the blood is flowing.

We can measure things that are linked to the actual perfusion of the tissue and the density of the vessels in the tissue.

There's different ways we can do it.

We can inject the bubbles in as a bolus and watch them flood into a region of tissue.

Or we can set up an infusion where we reach a steady state in the body.

Then we use the ultrasound itself to destroy some of the bubbles and then watch them quickly flow in.

What we should remember is this isn't a meeting where cardiologists are welcome.

Our friends in cardiology have a lot of experience with this kind of quantitative use of contrast agents.

There's lots of tricks and tools that they have available to them that you should research and make sure you're aware of if you're thinking of doing some of these types of studies.

Tools for Quantitative Analysis

What kind of tools do we need?

You heard the first speaker mention that Bracco themselves, the company that makes SonoVue, which is the contrast agent you've heard most about today.

They make a tool that's useful for doing quantitative imaging, a software tool called ViewBox.

What does this tool do? It allows you to feed in the sequence of ultrasound image data that you have, and then you can draw regions of interest on it.

You can extract the underlying information in those images and form things called time intensity curves.

That's one form of software.

Many of the ultrasound scanners, the scanner manufacturers themselves provide similar types of quantitative research tools.

If you're in the position of Professor Du and you work in an academic university, you can find some poor soul like me and get him to write you software to do it for you for free too.

Software Functionality

What do these types of software have to do?

They have to take in the data.

They have to in some ways help you to correct the data for motion.

You see here on this slide, this funny set of images that I'm pointing to now show how just using the background B-mode data, we can compare the speckle pattern from one frame to the next.

Then we can try and adjust the motion of the scan so that we correct, so that we continue looking at the same region of tissue throughout the contrast study.

We need some way to be able to put on multiple regions of interest.

You can have local references.

You can have a control region and a region where you suspect your tumor is.

You want the software to be able to do some sort of data quality checking.

For example, if the patient coughs or the radiologist leans over to pick up their coffee and start scanning the wrong section, the image sequence will notice that it's gone wrong and won't suddenly confound your data by including that into the analysis.

Often putting all this together for a specific clinical application requires some kind of bespoke solution.

Something out of a box won't quite do what you want it to do.

Having a direct link to either the manufacturers or to an academic engineer like myself who can tweak the software to do what you wanna do is probably really useful in these kinds of circumstances.

At the end of the day, what you get is a graph which has time on the x axis and some measure of brightness, some measure of intensity or brightness on the vertical axis.

Fundamental Assumptions in Measurement

This is where we just take a moment to think about what we're measuring, what we want to measure.

We think we've got this fundamental assumption that the microbubbles stay in the blood.

The microbubbles represent blood.

What we're trying to measure is how much blood, how many blood vessels or how much blood is there at any one moment in time, and how that changes over time.

The fundamental assumption is that the microbubbles represent blood.

The next assumption is that the information on our images, the brightness of our images represents microbubbles.

These two, this chain of assumptions are really important in this quantitative research.

We're making a measurement of the concentration of microbubbles in a region of your image.

We make that measurement at a point in time, and we can track it over time using the sequence of images that we collect.

You can imagine, if this was a classroom full of students, I'd probably stop you now and ask you all to give me suggestions for how this could go wrong.

I'm sure you could come up with an endless list, which is what I did with my colleagues a few years ago.

The paper that was referenced a couple of slides ago, and I'll reference it again later on, does exactly that.

It's a paper of doom and gloom.

It's a paper that says, these are all the ways you can get it wrong.

Actually with not too many suggestions for how to get it right, but that was published a few years ago, and we've made a bit more of an inroad into sorting out some of these problems.

Potential Pitfalls in Quantitative Measurement

Broadly speaking, this is a kind of slightly artificial categorization, but you could categorize the issues into these broad patterns.

Many of the issues that you might come across are to do with your scanner.

A lot of these issues you can control, you can deal with, you just need a bit of foresight to make sure that you don't mess it up.

We'll talk about one or two of those in a minute.

Your patient or the environment that you put your bubbles into is gonna potentially confound this issue dramatically.

We have to be aware of some of these circumstances.

There are some things we can do, and in some cases, there's nothing we can do.

They will just cause error in our analysis at the end, and we just have to make sure that we know that that is why that error is there.

We have to design our study carefully enough that we can characterize that error.

The final category are the bubbles themselves.

There's different types of bubble.

The way we handle the bubbles will change their behavior.

The type of study we're doing, the length of the study and so on, depending on the types of bubbles will all have impact into the errors that might be associated with the measurements we get.

For each of these cases throughout the rest of my talk, I'm just gonna pick out one or two examples and just show you how they might be a problem and what you need to do about it.

Scanner-Related Issues

The first one, I won't name names here because some of these are embarrassing mistakes made by clinicians.

The first one, if you watch this sequence carefully, you'll notice halfway through it something gets brighter.

This is 'cause the clinician couldn't resist the temptation to change the gain on the scanner halfway through.

If I'm basing a measurement on the brightness of my ultrasound scan, and I'm assuming that the brightness of my ultrasound scan tells me how many bubbles are in the blood, the last thing you want is that brightness to change halfway through.

This is a good example of controls on the scanner that you have to design your study and you have to fix those controls because if you're gonna do quantitative analysis of the data afterwards, if this sneaked through, if you didn't notice this, it would confound your results.

That's a trivial one.

A really important one to plan in advance for perhaps less trivial is the complex relationship of the behavior of the bubble to the acoustic environment it finds itself in.

We have touched on this in a couple of talks right at the beginning of the day.

Bubbles are very strong scatters of the sound, and they are very interesting scatters of sound, and that they behave in this non-linear way, and they exhibit this resonance phenomena.

What do these two things mean?

These two graphs show you what these mean for us.

This graph here along the X axis is increasing power.

I'm basically turning up the mechanical index of my scanner.

Everything else is the same. I'm just turning up the output power of my scanner.

Along the vertical axis is a measurement of how much signal I'm getting back.

It's not a straight linear relationship.

There's a curve there. What this is telling me is that as I increase my power or my output power of my scanner, I get more back from my money.

That means that there's, since there's not a linear relationship there, if the bubbles in different regions of your image or in different times of your image, or if you change the power as you're scanning, you will get a nonlinear relationship in your quantitative measurement.

It won't make any sense.

The other graph has frequency along the x axis and amount of scattered signal you get back on the Y axis.

This basically just shows you how resonant the bubbles are.

These particular bubbles have a stronger resonance around two megahertz.

That means that if you are using a scanner which you can use at two megahertz, you'll get more signal back.

Whereas if you tweak that frequency or use the wrong probe, you'll get a lesser signal back.

But again, that would be nothing to do with the concentration of bubbles, which is what you are trying to measure.

It's all to do with the settings of your scanner.

These are the kind of things that with a bit of foreknowledge, you need to set up appropriately with a bit of knowledge of what the physics of how the bubble behaves.

You might be able to tune them so you get good signals.

Whatever you do, if you're trying to do a quantitative study and especially a longitudinal study where you're scanning different patients and things, you need to set up a protocol where these things end up being as fixed so that you can make comparisons.

Patient and Environment Factors

If we think of a couple of examples of how the patient or the environment that we put the bubbles into might make this a challenge.

If we have a scan of the liver, and we look at two regions of interest in the liver and plot these time intensity curves for those two regions, just suspending any disbelief for a moment.

If we assume that those two regions of the liver are the same, the same normal liver tissue, but they're at different depths in the body, then the overlying attenuation of the tissues will mean that you get a different measure from those two regions because one is deeper in the body than the other.

There's not a huge amount you can do for this.

I'm gonna show you on the next slide, one or two ways you might be able to correct this under specific circumstances.

But if you're not aware of this and you design these quantitative studies, then it's going to lead to errors in your final analysis that you don't understand.

One specific situation where you may be able to account for this is where you have a large area of, so this is a carotid artery, a nonlinear image from a carotid artery.

You can see something quite interesting on these images.

The overlying material.

Overlying tissue that you can't see in this cropped bit of image is causing some regions to be darker than other regions.

One thing we know about this, or we can assume about this image, is that the concentration of contrast agent in this large vessel is uniform at this moment in time.

We can say, if this bit's this bright, then this bit should be similarly bright too, and we can write some image processing tools, some signal processing tools to correct that.

If I flick back between those two, that's exactly what we did in our research.

We worked out, we worked on a simple image processing tool that said if we assume that the contrast concentration in these two regions is the same, we can wipe out the effect of shadowing across this vessel, and we can extend that to regions further into the tissue below the vessel to mean that if we're trying to make quantitative measurements of on the far side of the carotid artery, there'll be more likely to get realistic quantitative measurements.

Another aspect, which can be troublesome when the bubbles are in a patient in a realistic situation is that often there'll be, as the sound propagates through the bubbles, not only is the sound attenuated, we've already heard that mentioned today, but the sound changes its nature.

The bubbles are highly non-linear in the way they scatter the sound, but they also cause the sound that propagates through them to propagate in a non-linear way.

What this means is that when the sound propagates through a cloud of bubbles, when it gets to the other side, that sound wave now has nonlinear characteristics.

Then it scatters off linear tissue behind the vessel.

That sound that comes back to your transducer is interpreted by your scanner, the signal processing in your scanner as lots of bubbles, which weren't there at all.

It's just bright echoes from linear structures on the far side of a cloud of bubbles.

These are two examples of clinical situations where that's happened.

There's bubbles here, but this is just normal tissue on the far side, which is now brightened up because of this non-linear propagation artifact.

In the carotid, you can get artifacts where regions of tissue on the other side of your carotid artery will look artificially bright 'cause of this nonlinear propagation process.

In our research, we came up with some ways to correct for this.

This is just a phantom image with a vessel full of microbubbles at the top and then lots of tissue mimicking material below it, which shouldn't be lighting up in our non-linear image here.

We've worked out how to correct that.

The way we correct that is basically uses the fact that we can collect a linear image and a non-linear image at the same time.

We can in a relatively iterative way, we can look at what's in the linear image, look at what's in the non-linear image and say, there's very strong echoes in that linear image.

They're also existent in the non-linear image.

We can be pretty sure they come from the fact that those linear structures are breaking through into the nonlinear image and we can correct for that artifact.

Here's an example where it's perhaps a bit more exciting in that this is a phantom in which we, so this is a standard linear mode, linear mode in B mode image of our phantom.

This is a vessel. Then there's tissue mimicking structure below that and hidden in that tissue mimicking structure, there's a very small, millimeter sized vascular vessel structure.

When we do our non-linear image, our vessel lights up and the tissue has this non-linear artifact, and we can't see that tiny sub millimeter vessel in the deeper part.

But when we add our correction algorithm, suddenly the non-linear artifact is reduced and we can reveal that vessel from within that original artifact.

This is an example of how, and a few years ago when I gave this talk, this was an artifact that we had no idea how to correct.

This is an example of one of those things that used to be doom and gloom.

Now we're thinking of, we've come up with some ways that we may be able to correct for this and incorporate into future studies.

Bubble-Related Factors

This is a slightly different thing.

When you start messing around with microbubbles and thinking about doing quantitative studies, quite often you're tempted to do something in the lab.

In basic science in the lab away from your patients, you think, I just need to check, for example, the dose response of this.

I need to put bubbles in a beaker, change the dose and see if my scanner really does have a linear relationship with how many bubbles I put in.

We did stuff like this for years.

Then one day we suddenly thought we should check something a bit naively.

We'd been doing these experiments just at the room temperature of the lab, and we thought we should really check to see how differently the bubbles behaved at different temperatures.

These are images of bubbles taken using a high speed microscopic camera.

It's taking millions of frames a second of a bubble actually oscillating in the sound field.

This is a relatively large bubble, about 10 microns in diameter.

We found, we did this over hundreds of bubbles, but this is just one example.

These are two bubbles that appear to be about the same size.

We've done two experiments, one where the bubble was in a beaker at room temperature, and one where we just heated the beaker up to room temperature before we did the experiment.

You can see that the oscillations of the bubble that's warmer are much more dramatic than the oscillations of the bubble that's at room temperature. So this is another case where the environment that you put the bubble into could affect the response that you get back from the bubble and could mess up your quantitation.

If we move on to the final section, which is the bubbles themselves, and how you handle the bubbles.

Here's a pretty obvious one I guess.

If you're giving a bolus injection and measuring how that contrast agent sweeps in, you are measuring one of these time intensity curves where time is along the x axis and the intensity is along the y axis, and you give a short sharp bolus, you can imagine that you get a short sharp rise and then a washout.

Whereas if you do that injection slower, you will get a different shaped curve.

That's something that you need to think about when you're designing these kind of quantitative studies that you have a consistent way for delivering the same kind of bolus each time.

Otherwise, your results will be confounded by the effect of the injection rate.

Here's something that we found in our lab that other people have found recently as well, is that we've already discussed how different bubbles, the bubbles have a different resonance depending on their size.

But what people have found recently as well is that even two bubbles, which are exactly the same size, can have a very different acoustic response.

These two bubbles apparently the same size.

They're the same type of bubble. They're both SonoVue, they're both being excited by exactly the same sound field, but one of them is doing something quite different to the other.

What people are now discovering is with many of these bubbles, which are made by a lipid surfactant material over the bubble, is that the bubbles might appear to be the same size under a microscope, but they have a different concentration of structural material on their shell.

Therefore, they might behave differently.

There's nothing much you can do about this in your research, but it's an example of how the signal you get might not reflect the concentration of bubbles.

It is an example of why sometimes there's some things that we need to know more about before we can really hone in on some of the quantitative measurements.

Summary

To summarize, these two days are all about showing you the power of contrast agents.

My talk is showing you really or trying to help you to realize that quantitative measurement for these kind of studies is really important.

It's a really important way of establishing the use of microbubbles.

It's really important for finding new uses for microbubbles.

The challenges of making quantitative measurements with bubbles are high.

There are many ways that the answers you get could be confounded.

If we don't take those into account, what happens?

We get variable results.

We get errors that we don't understand in our quantitative studies.

What does that do? It messes up the whole field.

It means that it's harder to get things licensed.

It's harder to prove things in the literature.

It's harder to get NICE to accept the guidelines.

We need to be consistent when we do our quantitative research so that we can push the field forward.

Most of the factors that confound quantitative measurement in ultrasound can definitely be dealt with.

We can characterize them.

We have characterized them, we've written papers on their effects, but we can definitely design studies where we can control for them or we can even eliminate them.

That's the kind of thing that you need to just have in your mind when you're designing these studies.

Of course there's a few, and the new ones come up all the time.

There's some features of bubbles that we haven't quite understood yet and could be confounding our results.

We're gradually working our way through these to make it a more quantitative field.

There's many people involved in this research that I should acknowledge.

I won't go through them all one by one, but also my funding mainly from the EPSRC and also the Wellcome Trust.

Thank you very much for your attention.