Educational content on VJOncology is intended for healthcare professionals only. By visiting this website and accessing this information you confirm that you are a healthcare professional.

Share this video  

WCLC Sept 2021 | Radiomics in oncology

Dirk De Ruysscher, MD, PhD, Maastricht University Medical Center, Maastricht, Netherlands, discusses the field of radiomics and highlights the application of radiomics in oncology. Radiomics uses large numbers of quantitative features from images, for example taken from CT or MRI scans, and extracts these features mathematically. Radiomics has many applications in oncology, including prognostication, predictive testing and response assessments. Radiomics has the potential to make a big impact in oncology; however, further studies are required for further validation and reproducibility. This interview took place at the World Conference on Lung Cancer (WCLC) 2021.

Transcript (edited for clarity)

Radiomics actually is a quantitative image analysis. So basically, in any image, whether it’s a CT scan, or an MRI scan, or the picture of you and me at this moment, actually what we see are voxels, or two dimensional structures. And what is in a picture is actually, there’s much more information than what the human eye can see. And that is logical. We all know that, that the human eye has some limitations, of course...

Radiomics actually is a quantitative image analysis. So basically, in any image, whether it’s a CT scan, or an MRI scan, or the picture of you and me at this moment, actually what we see are voxels, or two dimensional structures. And what is in a picture is actually, there’s much more information than what the human eye can see. And that is logical. We all know that, that the human eye has some limitations, of course. For instance, a television screen, the more pixels you have, the sharper your images, but there’s also interaction between pixels, and that quantitative interaction is called radiomics. So what we do, actually, we take into account every information, what is available in a certain picture. And we relate that type of information with outcome of patients. That was actually the beginning of radiomics, and that is still an area of ongoing research.

So that means in practice, that when we see, for instance, a lung cancer on a CT scan, we would say, well, on that image, we only see a quite homogeneous mass. Maybe there’s some density changes within the same tumor mass. But in reality, when we bring this all together, how the shape is, how all the texture is, the granularity, and everything in the image, we can say this is due to some genetic features or other features of the tumor, like hypoxia. All this is also logical if you think about when I see you and I see you have brown eyes, for instance, that is actually because there’s some genetic features, which makes you to have brown eyes, for instance. And that is exactly the same, what we do.

But by looking this in a purely mathematical way, you can make this kind of quantitative analysis very objective. So it is not depending on the observer, it’s not depending on the experience of the doctor, or whoever, you can really make this really mathematically very sound. And what has been done at this moment by many groups, including our own group, is actually that by this type of quantitative image analysis, you can relate this to the prognosis of patients. So that means that indeed, there’s relation between some characteristics on the image and the survival of lung cancer patients, of colorectal cancer patients. That’s one.

Second, you can also use this type of analysis as a more sensitive way, compared to volumetric changes, in order to relate to the response of patients for systemic treatment. As an example, for instance, what in general is done, you give a patient immune treatment, and when the tumor is getting smaller, we say this is a response. And that’s very good, but of course, we know that before the tumor shrinks, there’s a lot of biology going on, and we can capture this with a certain extent with radiomics.

And lastly, what we can also do with radiomics, that is we can relate that image or those features with biology. For instance, infiltration of CD8 T-cells, which are the effective cells for immune treatment. So although you may say, “Well, we are only looking at an image”, there’s also biology, and there’s a correlation. And that is something, what is very interesting, of course, because this means that without adding new types of imaging, and that is really important because we don’t need another CT scan or MRI scan or whatever, by using the existing images, we can improve the prognostic and the predictive value of our images. And of course, we should bring, and we are bringing, this together with known biology and with known artificial intelligence. And this will lead, I’m sure, within a couple of years, to really very practical applications which can be broadly used, because the only thing what you need is the appropriate software, and you can use existing images.

And we can use radiomics together with artificial intelligence. There is some differences between the two, but more and more, they are merging. And we can use this also to make the whole workflow of radiation oncology, and oncology in general, more efficient. For instance, we can use those automatic processes to define tumors, lymph nodes, organ risk, like the heart, the lungs, and so on, before the treatment, but also during the treatment on the daily cone-beam CT scans, in order to know when we should do adaptation, how we should do it in an automatic workflow. Because this automatic workflow will, and this is partly implemented already in clinical practice, we can use this to make the whole process more robust. So that means less inter-observer variability, and at the same time, less costly. Because, of course, costs are a very, very major issue in every country. So all of this can be used actually, where radiomics is a part of.

So I would say, from a biological point of view, prognostic, predictive, and on the other point, from the work flow point of view. And part of this is reality, and part of this is work in progress. And I think, the major threat in radiomics at this moment, that is that it’s so easy to use some applications, software applications, for radiomics, but the quality of a lot of publications is pretty low. And what we should do now is really invest in very high quality data, including reproducibility, that is a major issue there. Including to see which type of image you can use, how you should use it, and so on, in order to really make this an applicable application. But I’m pretty convinced that it will be possible, and there are already some first applications which can be used in daily practice. So I think, it’s a matter of improving the quality, and then it will be implemented in clinical practice for some applications.

Read more...