Background

“One-size-fits-all” is one of the major obstacles to effective cancer treatment in modern medicine. Personalized cancer treatment must consider regional genomic diversity of the tumor. Different regions of the same tumor, or the primary and metastatic tumors, can have widely variable genetic signatures. This diversity causes the tumor to respond inconsistently to targeted therapy, resulting in proliferation of resistant clones and recurrence of the tumor. Mapping the genomic diversity within each tumor could improve clinical outcomes by providing information for the selection of mutations-based combinatorial therapies.

Characterization of global tumor genomics is very challenging. The existing biopsy-based genetic profiling is invasive, costly, time-consuming, and not part of routine clinical care. Therefore, it can only be done for a few locations within a tumor. This fails to provide global tumor genomic assessment. On the other hand, imaging techniques, such as CT, MRI, and PET, can non-invasively and fast scan the entire tumor in routine clinical care. While it has been traditionally thought that imaging lacks enough resolution to determine the genomics of a tumor, new advances in imaging techniques and new studies support the hypothesis that imaging can reflect the underlying tumor genotypes. If one could identify an imagenomics association map (i.e., imaging features that accurately predict genomic signatures) it would then be possible to provide imaging-based non-invasive genomic characterization across the entire tumor. The challenges of this research include development of novel statistics and machine learning methods to integrate two sources of large data—(1) multi-type genomic data (expression, aCGH, exon) and (2) multi-modality imaging data (multi-parametric MRIs, CT, DECT, PET). The challenges of this research include development of novel statistics and machine learning methods to integrate two sources of big data—multi-type genomic data (expression, aCGH, exon) and multi-modality imaging data (multi-parametric MRIs, CT, DE-CT, PET)—as well as fulfillment of clinical standards in sensitivity, specificity, reliability and reproducibility for translational research.

Objective

The objective of this research is to identify an imagenomics association map to enable imaging-based genomic characterization of a tumor across different spatial locations of the tumor. The ultimate goal is to identify imaging-based genomic biomarkers to bring about effective treatment that is optimized for regional tumor genomic diversity.

Approach

The research objective is accomplished through novel big data analytics and machine learning such as sparse learning and transfer learning, as well as solid clinical validation.

Impact

Our expanding knowledge of the genetic basis and molecular mechanisms of cancer is beginning to revolutionize clinical cancer practice. Personalized medicine (using biomarkers to classify tumors and direct treatment decisions profiles) is becoming the new standard of care. Identification of molecular/genomic biomarkers has led to changes in treatment, such as new, targeted therapies aimed at specific tumor receptors or molecular markers. This research will contribute to personalized cancer medicine by providing non-invasive imaging-based biomarkers capable of providing information about tumor genomic diversity and eventually leading to better treatment and clinical outcomes.

Figure 1: imagenomic research roadmap.

Source: R21NS082609-02