Click here to

Session: Imaging for Treatment Assessment and Outcome Modeling [Return to Session]

CT Radiomics Analysis On Whole Pancreas Between Healthy Individual and PDAC Patients: Uncertainty Analysis and Predictive Modeling

A Kolomaya1, G Ostdiek-wille2, J Wong3, X Cheng4, C Lin5, Y Lei6, S Zhou7, S Wang8*, (1) University of Nebraska Medical Center, ,,(2) University of Nebraska Medical Center, ,,(3) University of Nebraska Medical Center, Omaha, NE, (4) University of Nebraska Omaha, ,,(5) University of Nebraska Medical Center, Omaha, NE, (6) Barrow Neurological Institute, Phoenix, AZ, (7) University of Nebraska Medical Center, Omaha, NE, (8) University of Nebraska Medical Center, Omaha, NE


SU-F-206-4 (Sunday, 7/10/2022) 2:00 PM - 3:00 PM [Eastern Time (GMT-4)]

Room 206

Purpose: In this study, we analyzed the uncertainty of the radiomics features of the whole pancreas between healthy individuals and pancreatic cancer patients. We, then, established a predictive model that can distinguish cancer patients from healthy individuals based on imaging features.

Methods: We retrospectively collected venous phase of contrast CT images from 181 control subjects and 85 cancer case subjects for radiomics analysis and predictive modeling. An attending radiation oncologist manually delineated the pancreas for all the subjects using the Varian Eclipse treatment planning system, and we extracted 924 features in the PyRadiomics platform. We randomly selected 188 cases (including 60 cancer cases and 129 control cases) as the training set and used the rest of the data, 25 cancer cases and 52 control cases, as a test set. We established a feature selection pipeline to rule out redundant or unstable features. Nine features were determined to be stable and non-redundant, and we trained a Random Forest model utilizing these nine features to distinguish the cancer patients from the healthy individuals on the training dataset. We analyzed the performance of our nine-feature model by running 5-fold cross-validation on the training dataset. At last, we applied our best model to the test set (52 normal and 25 cancer).

Results: We identified that 311 radiomics features are stable against various uncertainty sources, including bin width, resampling, image transformation, image noise, and segmentation uncertainty. Our predictive model has achieved a mean AUC of 0.99 ± 0.01 on the training dataset (189 subjects) by cross-validation. The model achieved an AUC of 0.901 on the independent test set (77 subjects) and an accuracy of 0.922

Conclusion: Radiomics analysis based on whole pancreas is a promising tool to distinguish cancer patients from healthy individuals, and potentially becomes an early detection tool for pancreatic cancer.

Funding Support, Disclosures, and Conflict of Interest: This work is partially supported by Otis-Glebe Medical Foundation


Not Applicable / None Entered.


Not Applicable / None Entered.

Contact Email