Oncologists’ Guide to Genomics

Stay current on the latest trends in genomics and molecular diagnostics for oncology.

Tuesday, March 20, 2018

Radiomics-Based Imaging Tool May Predict Response to Immunotherapy

By Catlin Nalley

"Immunotherapy has profoundly changed the management of multiple cancers," said Roger Sun, MD, PhD candidate under Eric Deutsch, MD, PhD, and Charles Ferté, MD, PhD, at the laboratory INSERM U1030 at Gustave Roussy in Villejuif, France. "However, most patients do not respond to this type of treatment. That is why we need to identify biomarkers that allow identification of patients who are most likely to respond to immunotherapy."

Studies utilizing biopsy samples of tumor tissues have confirmed the link between immune-cell infiltration into tumors and patients' treatment responses; however, Sun noted, because cancers are heterogeneous, biopsies only reflect the local aspect of the tumor.

"Medical computational imaging, also known as radiomics, is a new field of research that aims to translate standard imaging like CT, MRI, or PET into objective data and use them as biomarkers," he explained.

New data, presented at the AACR-NCI-EORTC International Conference on Molecular Targets and Cancer Therapeutics, suggests a computational imaging-based signature of immune-cell infiltration in and around a tumor could predict patients' responses to treatment with anti-PD-1/PD-L1 immunotherapies (Abstract A051).

Study Details

The research, spearheaded by Deutsch, Head of the INSERM U1030 Research Unit at Gustave Roussy, and Ferté, a medical oncologist and computational biologist from Gustave Roussy, aimed to develop a radiomics-based imaging tool of tumor immune infiltrate and assess whether this tool could predict clinical outcomes of patients treated with anti-PD-1/PD-L1 therapy.

"We wanted to explore the potential of radiomics in the other emerging field of immunotherapy, for which there are no established biomarkers to predict patient responses," study author Sun told Oncology Times.

"Moreover, this imaging biomarker is non-invasive, cost-effective, can be applied on all the tumor localizations [to] evaluate the tumor heterogeneity, and can be repeated throughout the course of disease," he emphasized. "Development of such a tool would be of highest interest in improving patient care."

The researchers utilized radiomics to estimate the abundance of immune-cell infiltration in tumors and assess their potential to predict response to anti-PD-1/PD-L1 therapies. Data from the head and neck, liver, lung, and bladder cohorts of The Cancer Imaging Archive was used by the team to develop a radiomics-based model of tumor-infiltrating effector T cells (Teff).

Eighty radiomics features were extracted, according to investigators, and a radiomics score was built that could predict the abundance of tumor-infiltrating Teff estimated using RNA sequencing data.

"As tumor inflammation is known to be related with clinical response to immunotherapy, we used genomic data of tumors from The Cancer Genome Atlas to quantify tumor inflammation using a published gene expression signature," Sun explained. "We then trained a radiomic signature based on the CT images of the corresponding tumor using machine learning approaches to predict tumor inflammation."

The radiomics score was initially tested on the CT scans of a cohort of 134 patients who had RNA-seq data available, investigators reported. Data showed that the radiomics score of Teff correlated with the genomics-based score of Teff.

"We validated this radiomic signature in a prospective cohort from our center, the MOSCATO trial, a precision medicine program where genomic information was extracted upon CT-guided biopsy," Sun noted. "For these patients, both contrast-enhanced CT scan at the time of the biopsy and RNA-seq data were also available.

"A third cohort of patients treated with immunotherapy was used to assess the association between the radiomic signature and outcome of patients (overall survival)."

The radiomics score was applied on data from the entire cohort and used the median value to divide the cohort into two groups: patients whose scores were above the median and those whose scores were below the median. According to investigators, "at any given time point, patients with a high score were 1.5 times more likely to be alive compared with those who had a low score."

"A signature based on imaging features, learned on genomic data, can reflect the tumor phenotype (low or high level of immune cells) and the response to immunotherapy," Sun said.

However, he acknowledged there are research limitations to consider, "Cohorts of patients were heterogeneous, with different types of tumors and varying imaging protocols, which could have impacted the radiomic signature.

"Even so, this data reflects the quality of data one can expect in 'real life,'" he continued. "Moreover, having access to the tumors' genomic data and the corresponding images in a large cohort is seldom possible, hence the data obtained is very valuable and is a unique opportunity to assess whether the radiomics data could provide an estimate of immune infiltration (assessed by RNA-seq)."

Practice Implications

This study is a positive step forward to better understand the potential clinical implications of radiomics.

"We are very encouraged by our findings that a signature based on imaging features could reflect the tumor immune infiltration and could predict response to immunotherapy," Sun noted. "These results are preliminary, and we need further clinical studies to validate them.

"Ultimately, this score may be useful to drive immunotherapy trials allowing stratification of patients. For radiotherapy-immunotherapy combinations, this score can be useful to identify which lesions to irradiate, in order to get some abscopal effects," he continued. "Enhancing data sharing and facilitating patient recruitment in clinical trials are necessary. With further improvements to this field with multi-disciplinary working groups, radiomics can become a reliable part of the decision support system in oncology."

Image computing can help extract information of the molecular and cellular composition of tissues, according to Sun. "Given the vast amount of medical images generated in oncology, the potential of radiomics in oncology is very huge and efforts have to be made to develop this very promising field of research," he concluded.

Catlin Nalley is associate editor.