Using Quantitative Radiomics and Cyst Fluid Biomarkers to Predict Pancreatic Cyst Type
Sophia Hernandez, BS, Andre Luiz Lourenco, PhD, Ezgi Kirimli, MD, Tyler J. York, MD, Alexa Glencer, MD, Patricia C. Conroy, MD, Charles S. Craik, PhD, Eugene J. Koay, MD, PhD, Spencer C. Behr, MD, Kimberly Kirkwood, MD
Introduction: Mucinous pancreatic cystic lesions (PCLs) have malignant potential while nonmucinous PCLs have minimal/no malignant potential. Accurate identification of nonmucinous PCLs would allow ∼25% of patients to potentially avoid further invasive testing or surgery. We used a combination of radiomic features and cyst fluid biomarkers (CEA and the protease Gastricsin) to identify mucinous PCLs.
Methods: We performed a retrospective analysis of preprocedural pancreatic protocol CT images and cyst fluid CEA/Gastricsin levels obtained from patients with pathologically confirmed PCLs from 2016-2019. We contoured images using 3D Slicer and extracted radiomic features using IBEX software. Three-fold, cross-validated, random forest models were created to produce ROC curves, and measure sensitivity, specificity, positive predictive values, and negative predictive values for identification of mucinous PCLs.
Results: CT images and cyst fluid from 41 patients with mucinous (n=31) and nonmucinous (n=10) PCLs were analyzed. Among 157 radiomic features, the 3 best were used, and together this model predicted mucinous status with an area under the curve (AUC) =0.73 and accuracy=83%. CEA alone was a poor predictor (n=29)(AUC 0.64; accuracy=52%). The addition of cyst fluid biomarkers improved the radiomic feature model accuracy. Adding CEA to the radiomic model increased the sensitivity, and adding cyst fluid Gastricsin to the radiomic model improved the accuracy and predictive ability (AUC 0.98; accuracy=92%) (Table 1).
Conclusion: Quantitative radiomics is a promising technique for noninvasive identification of mucinous PCLs that may be further enriched by cyst fluid analysis. These tools are expected to improve identification of patients who can be excluded from further invasive procedures and thereby avoid burdensome overtreatment.
Read Full Article Here: https://doi.org/10.1016/j.jamcollsurg.2021.07.307