Differentiating pancreatic mucinous cystic neoplasms form serous oligocystic adenomas in spectral CT images using machine learning algorithms: A preliminary study

Pancreatic cancer is one of the most fatal cancers. Distinguishing mucinous cystic neoplasm from serous oligocystic adenoma by using cross-sectional imaging system is very important for patients' prognosis. Gemstone spectral computed tomography (CT) can provide more information as compared with the conventional CT. Machine-learning algorithms have been employed in a great variety of applications. This preliminary study aims to verify the effectiveness of the additional information provided by spectral CT with the use of the state-of-the-art classification algorithm combined with feature-selection methods. Results show that SVM+MI achieves the highest classification accuracy (71.43%). The second highest classification accuracy is obtained by using SVM+LO (63.83%). Features selected by these algorithms are consistent with clinical observations. Top-ranking features include lower viewing energy (around 50 keV) CT values, Iodine-Water concentrations, and Effective-Z.