2.4. Statistical Analysis

The extracted radiomic margin characteristic values were normalized to be between 0 and 1. To test the reliability of extracted radiomic margin characteristics, the intraclass correlation coefficient (ICC) values were calculated by two radiologists.

For the prediction of survival, overall survival (OS) was defined as the time interval between the date of surgical resection and the date of death or last follow-up. The large number of radiomic margin characteristics was reduced using the least absolute shrinkage and selection operator (LASSO) method. A comparison between a stepwise multivariate logistic regression model with clinical features only, and a model integrating both clinical features and selected radiomic margin characteristics, was performed, with the calculation of C-indices included. Next, the internal validation of our prediction model was performed using 10-fold cross-validation. Finally, external validation of our prediction model was performed in a different study group by using the Cox regression, and time-dependent area under the curve (AUC) was calculated. Due to the absence of several clinical variables in the external validation group, which was downloaded from an open data source, external validation was performed with commonly existing clinical variables and all radiomic margin characteristics.

For the prediction of the MP subtype, the large number of radiomic margin characteristics was reduced using the LASSO method. Using radiomics features only, a model predicting the MP pattern was generated using logistic regression and split data validation.

All statistical analyses were performed with Statistical Analysis System (SAS version 9.4; SAS Institute, Cary, NC, USA) and R (version 3.3.1; Vienna, Austria; http://www.R-project.org/) software (accessed on 23 March 2020). p < 0.05 indicated statistical significance.