Fingerprinting Cancer: How Radiomics and Genomics Are Mapping Tumor Heterogeneity Using ‘Big Data’ to Track Killer Habitats

Cancer is often characterized as a serial murderer that needs to be struck down with brute force. But what researchers are slowly coming to understand is that this idea might actually make individual cancers stronger. The newest research very clearly exemplifies how a single “most-wanted” sign is not going to work to identify the veritable army of characteristics that a single tumor might have. The new paradigm of cancer heterogeneity has only just emerged within the past few years and is yet to be assimilated into every day patient management. Here is a snapshot of what the fields of radiomics and genetic research are working on to outsmart cancer in all its complexity.

Smarter segmentation

One of the leading thought-leaders in radiomics is Philippe Lambin, MD, PhD, chair and medical director of the Maastro Clinic, a division of the School for Oncology and Developmental Biology at the Maastricht University Medical Center in Maastricht, The Netherlands. Lambin and his colleagues are revolutionizing the way radiation oncology leverages advanced imaging to generate 3D parametric maps of tumors in a way that characterizes cancer better than the standard biopsy.

“Doctors are not very good at delineating tumors,” remarks Lambin about how radiomics stands to update a very outdated methodology of segmentation. “Volume is rarely taken into account, which is wrong, volume is very important.”

Lambin is working on a project called ‘Click and Grow,’ also known as single-click ensemble segmentation (SCES), which utilizes a complex radiomics algorithm to characterize the heterogeneity of tumors, in this case lung tumors using CT data. His collaborator on the project is Robert J. Gillies, PhD, chair of the department of cancer imaging and metabolism at H. Lee Moffitt Cancer Center in Tampa, Fla. Gillies agrees with Lambin on the importance of volume in delineating cancer.

“If you do things manually and allow the radiologist to delineate, then you consider that your gold standard and it’s a very low bar, indeed,” says Gillies. “Radiologists very rarely agree with each other. If you build the right software algorithm, you can get agreement more often than not—more than 95 percent. We get the same answer every time.”

This is no standard-uptake value plot. Click and Grow churns out a 3D volumetric detail based on hundreds of preset features and imaging data provided by PET, CT and MR and other imaging modalities. The algorithm includes 13 descriptors of size, 14 ways to describe location, 12 descriptors of shape, three different ways to measure volume, 17 ways to describe histogram intensity and three different types of texture measurements. The algorithm also looks at co-occurrence matrices 17 ways, 30 different wavelets and a total of 125 different Law features. The result is a very finely delineated lung tumor segmentation. This represents an encouraging advance in lung-tumor imaging, but it is not limited to lung cancer. The same concept can be used for a variety of cancers, including breast, prostate and pancreatic.

The undertaking of this software has been spread out over several international institutions including Lambin’s group in The Netherlands as well as the Institute of Automation in Beijing and others in Germany and the U.K. as well as Moffitt, Stanford, Harvard, Cornell and Columbia University. The last five of these collaborate on a kind of crowd-sourced programming of cloud data to continue perfecting the software. It may be sold between institutions at some point in the future to fund other research projects.

Why is this important?

Powerful image analysis is a necessity to fully understand the true genetic and biological nature of a tumor. This can lead not only to less variability in interpretation and more accuracy in tumor mapping, but it also can help prevent cancer outgrowth based on genetic mutation. Just when you think you know a tumor, you don’t, really.

“Medical oncology still attempts to treat cancers as homogeneous systems and they are not,” said Gillies. “Some people who are leaders in the field have recognized and promoted the idea that tumors are heterogeneous and probably become so as they age. Cancers very rarely arise out of nothing into a full-blown cancer like Venus on the half-shell.”

Biopsy, the standard of care, may not accurately represent the entirety of a tumor and, more often than not, it doesn’t (European Journal of Cancer; 2012, 48, 441–446). Cancers develop over years, if not decades and pick up a variety of genetic mutations in different regions of the tumor along the way. A biopsy in one area could lend itself to clinical decisions that have a potent effect on one area of a tumor, but a deleterious effect on another area. Therapy selection based on standard biopsy could not only undermine the success of a patient’s treatment, but could be contributing to the drug-resistance of their cancer. This is based on the evolutionary development of tumors.

“We study tumors using evolutionary dynamics and it is clearly stated that the road of evolution increases with the strength and the selection,” explains Gillies. “The stronger the selection force, the more rapid will be the road of evolution. When you try to treat cancers for maximum kill, that’s a very strong selection force. The failure on the part of therapies is represented by the fact that you are just putting a very, very strong selection force on the tumors and strongly selecting for regions that have resistance to those therapies—multiple therapies.”

What to do

Without creating a sense of despair over the standard’s limitations, researchers are encouraging adoption of a non-invasive route. Hence, ‘Click and Grow’ and other algorithms that use standard medical imaging to do what biopsy cannot.

“We do profound analyses of standard of care images to identify how many different habitats there are within tumors with the expectation and belief that the genetically heterogeneous cells are going to exist in the same habitats. In other words, the heterogeneity has been established by different selection forces acting on the tumor at different times during its developing,” states Gillies. “These selection forces we can characterize when there is a habitat. We believe that we can image those habitats using multi-parametric imaging.”

By habitats he means continents, at least metaphorically, that cover a vast topography of different environs and their ecologies, from mountainside to desert to seashore. It is the same for tumor cells that develop under particular environmental cues and constraints. Tumors can be host to a multitude of different kinds of tumor cells based on this view.

Genetic heterogeneity

Another way to improve our understanding of tumor complexity is through case-by-case genetic sequencing. Adrian V. Lee, PhD, director of the Women’s Cancer Research Center at the University of Pittsburgh Cancer Institute and Magee-Womens Research Institute in Pittsburgh, Pa., has worked extensively in genomics research and mirrors others’ complaints about standard biopsy, but there’s another way to go, he says, and that is through “deep” sequencing. 

“If you crush up a whole piece of tissue and you see a thousand mutations, you don’t know where they come from. Were they all in one cell? Were they spread across a thousand cells? But now we can do sequencing, metabolomics, proteomics, and all in great detail,” Lee explains.

How recent is the idea of cancer heterogeneity? Lee says it could be as old as 30 to 40 years, but that the extent of it was beyond anyone’s comprehension. Modern sequencing within just the past three or four years is driving the new paradigm of tumor profiling. While the human genome may have once cost $10 million to sequence, it now only costs $1,000 to do the same in just a decade’s time. Most leading medical institutions, Lee notes, now participate in ongoing genomics research projects, but more needs to be done to comprehend the vast amounts of resulting data.

“Tumors require very sophisticated analytics,” notes Lee. “We need that in-house and we need that desperately in areas of genomics, for example. We know now that there are no two cancers that are alike. Cancers are like fingerprints. Every single one of them is different,” says Lee, who is a part of a “big data” project involving breast cancer research looking into genetic differences between premenopausal and postmenopausal women. The findings were pointing straight to longitudinal heterogeneity and the necessity of personalized medicine, a topic intrinsic to breast cancer research. “Breast cancer has a way of being at the forefront, because it has the most molecular knowledge. Nearly all of the personalized ideas came through breast cancer—through estrogen receptors and novel surgical procedures. It’s always led the way.”

Again, this is not limited to breast cancer, but a variety of cancers, basically anywhere that a biopsy can be done, which has recently been expanded with the innovation of liquid biopsy. This technique provides the opportunity to test changes in circulating DNA and RNA, not just solid tumor cells. Still, genetic sequencing is not enough. Radiomics is necessary to complete the picture. That is afforded through advanced medical imaging and analysis using PET, SPECT, CT, MRI, ultrasound and others, although the majority of research is done on standard equipment for greater access and study replication.

Adopting adaptive therapy

Once extensive sequencing has shown oncologists the way forward, patients’ respective genetic mutations are treated differently.

“We screen for a mutation and give a therapy,” explains Lee. “When a patient becomes resistant to that therapy, we can then find genetic mutations that caused the resistance and then we can change their therapy.”

“Many groups are now doing this. We are sequencing humans and providing ther apy based on the genetic sequence of the tumor. This is a reality.”

What are the limitations of sequencing? We know that biopsy is limited, but could sequencing be similarly limited? Who is to say whether the sequences dug up are the ones to watch, or just minor-leaguers caught in the crossfire? These are questions still being answered by emerging research.

Killing off the idea of a cure

Both Lee and Gillies present the argument that selective force leads to drug-resistance and problems with recurrence down the line. The latter even ventured to change not just the way tumors are approached and delineated and how therapies are selected, but also the method of aggressive shock-and-awe treatments.

“Another thing we can do is quit treating for maximal kill and get smarter about what we want our end point to be,” insists Gillies. “We want our end point to be that the patient lives a long and healthy life; not that we make them very sick and not that we select them for a cancer that is going to grow back and be untreatable. The ‘quest for a cure’ has done more damage than good.”

Big data troubles

With the advent of imaging analysis, high-tech auto-segmentation software and deep sequencing, the challenge of data storage and data migration grows exponentially. Lee says the opportunity to create data is now unlimited, but our ability to make sense of it and translate it into any real knowledge is still narrow. 

“There are huge volumes of data, but we have trouble storing them, moving them and analyzing them,” remarks Lee. “We need new methods of looking at relationships between mutations and doing causal discovery, where you sift through tons of data.”

While biopsy is still a necessity, Gillies says the standard can be improved by properly annotating samples. “We can do a better job of keeping track of where in the tumor the biopsies were taken and what tissue is being represented.”

Looking long term

Genomic mapping extends far beyond the borders of oncology. Lee says it is just a matter of time before routine genetic tests for diseases like diabetes become commonplace. This won’t be just cancer. It’s too soon to tell if radiomics could be a way to determine risk factors for diabetes. For cancer patients, radiomics and genomics will drive adaptive therapies in the future.

“We can be better informed before going in to treat the tumor and we can see what we’re up against. We also can identify interesting habitats for biopsy,” says Gillies. “We need to think ahead as to what resistance mechanisms might allow a tumor to escape a particular type of therapy and either pretreat with that or follow up our original therapy with secondary therapies that are going to prevent resistance outgrowth.”

With this new paradigm of heterogeneity and marriage between radiomics and genomics, a lot more tumors will find themselves in the line-up, ready to be recognized by the authorities.