Data Visualization for Decision Support in Radiomics

Abstract

Medical treatment for cancer is performed in thefollowing order: detection of the lesion, differential diagnosis, andtreatment. Radiomics can be considered as an artificial intelligence systemthat supports the second half of medical treatment. In radiomic studies, it isimportant to develop interactive tools for decision support because thetreatment is determined by a counterbalance between doctor’s discretion andpatient choice. The purpose of this study is to propose tools for visualanalytics in the estimation of 1p/19q codeletion of brain tumor. We collected81 MR images from the LGG-1p19q deletion database in the Cancer ImagingArchive. We calculated 740 radiomic features from the tumor region. Logisticregression with 6 radiomic features selected by Lasso was employed forestimating the presence or absence of 1p/19q codeletion. Multidimensionalscaling and nomogram were also used for visual analytics. Since we were able tovisualize the reason for determining the output of logistic regression, ourproposed methods are considered to be useful as a tool for the decision supportin treatment strategy.