Skip to content Skip to navigation

Radiogenomics Research

A multi-level integrated model to identify prognostic cancer subtypes using radiogenomics

A clinically defined subtype, such as glioblastoma (GBM), is heterogeneous in that patients show disparate response to treatment. GBM is the most malignant brain cancer in adults. Previously, subtypes defined using single modality molecular or imaging data in GBM remained to be variable. The Cancer Genome Atlas (TCGA), an effort by the National Cancer Institute, made available multi-level molecular and imaging data for GBM and other cancer types. Thus, there is a need to develop methods to integrate a combination of molecular and quantitative imaging data to define subtypes that correlate with survival. To bridge the gap between genotypic level from molecular data and phenotypic level from imaging data, we applied the strategy of molecular analysis focusing on pathway analysis to find differentially dysregulated pathways and the imaging approach to use functional quantitative imaging information (location and perfusion features) to produce the integrated model. We currently work on 1) developing imaging analysis pipeline to extract tumor location features, 2) processing perfusion imaging to measure abnormal blood flow in the tumors, and 3) integrating imaging features into a pathway model. We have found some preliminary results for tumor locations associated with poor survival for GBM. Patients were stratified into poor and good survival groups (Figure 1).

The frequency of tumor occurrence across patients in each survival group was visualized as heatmaps (Figure 2), in which common events are shown as “hot” areas and less frequent events are shown as “cooler”. The heatmaps revealed that the poor and good survival groups have distinct imaging location phenotypes: the poor survival group of patients had tumors in the right deep white matter adjacent to the posterior lateral ventricle and its subventricular zone; in contrast, tumors associated with good survival were more diffusely distributed throughout the brain and did not localize to any particular anatomic region. Image analysis found consistent results: tumors in the right deep white matter were significantly associated with poor survival. Molecular characterization of this group of poor prognostic tumors had up-regulations of genes in the PDGFRA pathway.

Perfusion imaging and integrated analysis work is ongoing. Perfusion imaging is a type of functional magnetic resonance imaging (MR) modality that measures blood flow in the brain. Tumor growth requires vascular supply, and thus new blood vessels were formed from existing vasculature, a process known as angiogenesis. Tumor growth may also lead to blood brain barrier breakdown, which is often associated with abnormal level of blood flow in the brain. We are interested in identifying perfusion image features that are associated with prognosis (Figure 3), as well as gaining an understanding of the underlying biology of micro-vascular activity. Integrating imaging features with molecular features using a pathway model not only allows us to create a comprehensive profile of the disease, but also enables us to link phenotypic information with genomic information at the pathway level.