A major reason for gradually moving toward a precision medicine approach is the inability of existing clinical trials to deal with the huge amount of variation across cancers, their microenvironments, and individual patients.
“Clinical trials that identify effectiveness of treatments are applicable to patients who share those exact characteristics of the cohort who were studied in the clinical trial, but real-world patients differ in many different ways. Therefore, the treatments don’t work equally well in patients,” said Dr Rubin. “It would be too expensive to do [a] clinical trial including all possible combinations of features of patients.”
That’s where radiomics and pathomics come in. “It’s not as powerful evidence as a controlled clinical trial, but it’s better than the current situation, where there’s no evidence” for certain small subpopulations, noted Dr Rubin — so individual physicians must rely on heterogenous clinical trial findings and their own clinical judgment.
That doesn’t mean PD-L1 expression, TMB, and other biomarkers in current clinical use become obsolete. Rather, they become part of the overall input that goes into creating a noninvasive digital biopsy.
“The applications of these systems could vary according to our needs,” Dr Banna said. “If we need, for instance, to develop a system that is able to predict the likelihood of the presence of PD-L1 expression, or EGFR/ALK/ROS1 alterations, or high TMB, to optimize tumor sample processing, we could train the neural networks to do it. If instead we want to get an individual predictive signature, we could also put together information from different sources, such as radiological, pathological, and lab information, to create a specific profile to be validated.”
Sometimes all 3 data types — genomics, radiomics, and pathomics — might contribute to a digital biopsy, while other times, only 1 or 2 values may be needed, Rivka Colen, MD, assistant professor of cancer systems imaging at The University of Texas MD Anderson Cancer Center in Houston, told Cancer Therapy Advisor. She offered the example of a clinical trial for a specific mutation where not all patients respond.
“It is then that, for example, radiomics might be the next -omic information needed to stratify those patients — with that mutation and with a specific radiomic signature that predicts response — into clinical trials,” Dr Colen said. Subsequently, “patients [who are] likely [to] benefit from that specific therapy receive the treatment, and those unlikely to respond are offered an alternative therapy, to which they are more likely to respond.”
Dr Colen’s research has also revealed how radiomics may introduce opportunities for co-occurring preclinical research (on specific therapies or disease states) to be conducted.
“We found that humans, when matched with animal signatures with similar genomic profiles, demonstrated similar radiomic signatures; those with distinct genomic profiles had distinct radiomic signatures,” Dr Colen said. “This demonstrates that both animals and humans in cross-species can be followed in a similar fashion and used in co-clinical trials.”
Going Beyond Genomics
While genomics has dominated most discussions of precision medicine in cancer, there are aspects of malignancies that genomics alone cannot address.
“One of the examples I give my students is there is no gene that says you worked in the shipyard and inhaled asbestos fibers for 10 or 20 years and have mesothelioma,” Dr Prior said. “You can see that history in an image, but there’s no gene that’s going to tell you that.”
Then there’s the high level of overall variability in cancer that Dr Prior believes has yet to be fully appreciated.
“I keep coming back to this idea that certain tumors have their own microbial community, their own microbiome, which impacts how they relate to the stroma around them, their own [tissue] and the normal tissue microenvironment, and the relationship with the immune system,” he said. “Certain cancer types have this interesting relationship with their microenvironment and with the immune system such that they learn how to trick the immune system, and in response to the therapy, they learn even more. So the cancer keeps adapting to its environment and can reprogram the immune system.”
With so many factors influencing how a particular cancer behaves, the need for more than genomic information becomes clear. “The real value of radiomics is to help us understand and identify multiparametric phenotypes that have higher predictive value,” Dr Prior said.
Dr Banna and his colleagues reviewed some early examples of this research, such as a group who retrospectively used several cohorts — for a total of 491 patients — to develop a radiomic signature for tumor-infiltrating CD8 cells.4 Another study, in which Dr Rubin was senior author, used a larger data set of 2186 histopathological images from patients with non-small cell lung cancer to create a survival prediction model, which was then validated with 294 more images.5 Though Dr Banna’s team cited other examples as well, none of the studies used more than a couple thousand images — far fewer than what will be needed before clinical application is even a possibility.