Using ML-Based Virtual Metastasis Biopsy to Predict Colon Cancer Tumor Progression
For patients with colon cancer, current cancer surveillance and tissue sampling techniques are unable to capture tumor heterogeneity. Liquid biopsy approaches analyzing cell-free DNA (cfDNA) in the blood have been recently introduced as a minimally-invasive surrogate for detecting tumor heterogeneity and resistance to therapy. A major limitation of cfDNA is its global nature that is unable to resolve genetic alterations on a per-lesion basis. Imaging techniques that complement this approach are needed to better guide targeted treatment decisions.
In her 2018 GE Healthcare/RSNA Resident Research Grant project, Dania Daye, MD, PhD, is identifying quantitative MRI-based measures of intra-tumor heterogeneity as early predictors of lesion progression potential in patients with metastatic colorectal cancer and to correlate the imaging phenotypic signatures with cfDNA profiles.
Dr. Daye and her research team will use quantitative radiomics and machine learning (ML) techniques to define a heterogeneity-based tumor phenotypic imaging signature that is predictive of tumor progression. By examining the association between this phenotypic imaging signature and cfDNA results using regression analysis, they can assess the ability to predict KRAS mutation, the most common resistance mutation in colon cancer, using an ML model previously established for this purpose.
“MRI-based quantitative intra-tumor heterogeneity measures can serve as early predictors of tumor progression potential,” Dr. Daye said, “and may optimize prognosis prediction and treatment decisions in patients with metastatic colon cancer.”