The goal of this study was to measure the reproducibility of radiomic features in pancreatic parenchyma and ductal adenocarcinomas in patients who underwent consecutive contrast-enhanced computed tomography (CECT) scans. A majority of features were not reproducible (as defined by a concordance correlation coefficient of greater than 0.9) when comparing their values across consecutive CECTs obtained within a two week period. While preliminary, these results highlight the potential challenge of using single threshold values for individual radiomic features in clinical decision making. However, this study does not address the increasing use of a combination of radiomic features in a prediction model. The use of multiple radiomic features in a prediction model may not be hampered by the low reproducibility of individual features. Ultimately, validation of radiomics based prediction models requires external data sets, preferably in a multi-center setting.
Key points
- For pancreatic-derived radiomic features from contrast-enhanced CT (CECT), fewer than 25% are reproducible (with a threshold of CCC < 0.9) in a clinical heterogeneous dataset.
- Variations between CECT scans affected the number of reproducible radiomic features to a greater extent than variations in radiologist segmentation.
- A smaller number of pancreatic tumor-derived radiomic features were reproducible compared with pancreatic parenchyma-derived radiomic features under the same conditions.
Authors: Rikiya Yamashita, Thomas Perrin, Jayasree Chakraborty, Joanne F. Chou, Natally Horvat, Maura A. Koszalka, Abhishek Midya, Mithat Gonen, Peter Allen, William R. Jarnagin, Amber L. Simpson, Richard K.G. Do
