The 3-Dimensional-Computed Tomography Texture Is Useful to Predict Pancreatic Neuroendocrine Tumor Grading

The aim of this study is to evaluate the computed tomography texture parameters in predicting grading.

This study analyzed 68 nonfunctioning pancreatic neuroendocrine neoplasms (Pan-NENs). Clinical and radiological parameters were studied. Four model models were built, including clinical and standard radiologic parameters (model 1), first- and second-order computed tomography features (models 2 and 3), all parameters (model 4). The diagnostic accuracy was reported as area under the curve. A score was computed using the best model and validated to predict progression-free survival.

The size of tumors and heterogeneous enhancement were related to the risk of "non-G1" Pan-NENs (coefficients 0.471, P = 0.012, and 1.508, P = 0.027). Four second-order parameters were significantly related to the presence of "non-G1" Pan-NENs: the gray level co-occurrence matrix correlation (6.771; P = 0.011), gray level co-occurrence matrix contrast variance (0.349; P = 0.009), the neighborhood gray-level different matrix contrast (-63.129; P = 0.001), and the gray-level zone length matrix with the low gray-level zone emphasis (-0.151; P = 0.049). Model 4 was the best, with a higher area under the curve (0.912; P = 0.005). The score obtained predicted the progression-free survival.

Computed tomography radiomics signature can be useful in preoperative workup.