Sapporo Medical Journal

Abstract : The important role of positron emission tomography/computed tomography scans in oncology arises from the ability to track physiological changes, such as metabolism, within tumours. Nevertheless, a high metabolism rate in malignant lesions is known to be caused by activating other mutation pathways, such as the Kirsten rat sarcoma viral oncogene homolog (KRAS) and B-Raf proto-oncogene, serine/threonine kinase (BRAF) pathways, that induce other physiological alterations, such as angiogenesis, cell proliferation and differentiation. Objective: to explore whether KRAS/BRAF mutation pathways can be captured non-invasively by analysing the image features (radiomics) induced by F-18-fluorodeoxyglucose distribution. Methods: Fifty-six colorectal cancer cases were retrospectively collected. A semi-automated (GrowGut) algorithm was then applied for tumour delineation. Next, 106 features were extracted. A logistic regression model was trained after feature reduction via the Boruta algorithm. The classifier accuracy was then assessed by the area under the curve (AUC). Results: Long Run Emphasis, Short Run Emphasis and High Gray Level Zone Emphasis were found to be associated with KRAS/BRAF mutation. The model showed the highest accuracy (0.83) at the cut-off value of 0.51. The model yielded an AUC of 0.82 (95% CI, 0.69–0.94) with 63% sensitivity and 94% specificity. Conclusion: Our results suggest that KRAF/BRAF overexpression is associated with higher Short Run Emphasis and High G