Abstract No. 32 ▪ FEATURED ABSTRACT Predicting tumor grade and mutation status of pancreatic neuroendocrine liver metastases using computed tomography radiomic features

Abstract

Tumor grade and mutation status are putative biomarkers of response to trans-arterial embolization of pancreatic neuroendocrine liver metastases (PNLM). We sought to evaluate the effectiveness of using CT radiomic features to classify grade and mutation status of PNLM. 40 patients with PNLM who underwent CT scans and biopsies at a single center from April 2009 to February 2018 were included in the database. Biopsied tumors with recorded tumor grade (Grade 1-3) and mutation status (presence or absence of MEN1/DAXX/ATRX mutation) were segmented in both the hepatic arterial (HAP) and portal venous (PV) phases using 3D Slicer. Quantitative features (n = 115) from each scan were extracted from segmented tumors using PyRadiomics. Standard feature normalization was performed. Principal component analysis (PCA) was used for dimensionality reduction. Linear and radial basis function support vector machine (SVM) were used for training. K-fold cross-validation was used to estimate testing accuracy. Permutation testing was used to assess statistical significance. There were 25/40 (62.5%) patients with MEN1, DAXX, and/or ATRX mutation. There were 9/40 (22.5%) G1 tumors, 16/40 (40%) G2 tumors, and 37.5% G3 tumors. Over 95% of the variance in the data was described by the top 6 principal components, which were used for training and testing. Testing accuracy was 75% ± 11% for predicting mutation status (P = 0.008). Testing accuracy was 52.5% ± 11% for predicting tumor grade (P = 0.004). CT radiomic features can be used to predict mutation status and tumor grade in patients with PNLM.