Towards Personalized Oncology: Developing Bioimaging Markers for Cholangiocarcinoma

Saturday, 14 February 2015
Exhibit Hall (San Jose Convention Center)
Annelise M. Silva, Arizona State University, Tempe, AZ
Background: Cholangiocarcinoma is an intractable biliary malignancy with dismal long-term survival. Recent investigations have identified a driver oncogene, FGFR2-fusion, which responds to targeted chemotherapeutic agents. However, genomic analysis is invasive, time consuming, and incurs high costs. Thus, the purpose of this study was to utilize conventional and Dual Energy Computed Tomography (DECT) and Texture Analysis to develop a Radiogenomic method for identifying FGFR2-fusion. Methods: 16 patients with advanced sporadic intrahepatic cholangiocarcinoma who previously underwent integrated tumor sample whole genome/whole transcriptome analyses were evaluated with contrast enhanced DECT. Quantitative measurements were taken from regions of interest (ROIs) over representative areas of tumor and unaffected liver. On 140kVp arterial, portal venous and delayed phase images, min/max/mean/std.deviation Houndsfield units (HU) were measured. On delayed phase DECT (80/140 kVp) images, min/max/mean/std.deviation of effective-Z, material basis pairs (iodine, water, calcium), and spectral attenuation from 40-140 keV were measured. From this data, 42 independent variables were generated. ROIs were isolated using Hanning window prior to texture analysis (GLCM & FTFT). Statistical analysis performed: a) R to generate heat map data; b) Hierarchical Clustering (HC; XLSTAT 2013); and c).Random Tree (WEKA 3.6) classification models. Results: Blinded multivariate HC analysis showed large degree of dissimilarity between aggregate tumor data from patients with (n=3) versus without FGFR2-fusion (n=13). Classification decision tree analyses showed characteristics most predictive of FGFR2-fusion included mean arterial/venous enhancement, and mean effective-Z. The enhanced CT phases generated 84 GLCM & 81 FTFT variables. Random Tree classification models identified 28 FTFT features versus 14 GLCM features for accurate discrimination. For all models, delayed phase produced the most distinguishing features for FGFR2+. Conclusion: This study suggests that DECT imaging and advanced texture analysis may be useful in identification of a novel genomic target (FGFR2-fusion) in patients with advanced sporadic cholangiocarcinoma. A valid radiogenomic evaluation would allow for utilization of individualized gene-targeted therapy (via novel FGFR tyrosine kinase inhibitors in clinical development and FDA-approved) without need for multiple core biopsies, extended turn-around time (weeks to months) for sequencing, or prohibitively high expense associated with tissue-based genomic analysis.