A Nomogram for Early Prediction of Infected Pancreatic Necrosis Based on Contrast-Enhanced CT Radiomics and Inflammatory Indicators

Qing Yao,1 Yue Duan,2 Chao Jin,3 Xiang Li,3 Shiyu Wei,1 Yinghuan Shi,2 Yuelang Zhang,4 Jingyao Zhang,1 Chang Liu5 1Department of SICU, The First Affiliated Hospital of Xi’an Jiaotong University, Xi’an, 710061, People’s Republic of China; 2National Key Laboratory for Novel Software Technology, Nanjing University, Nanjing, 210008, People’s Republic of China; 3Department of Radiology, The First Affiliated Hospital of Xi’an Jiaotong University, Xi’an, 710061, People’s Republic of China; 4Department of Radiology, The Second Affiliated Hospital of Xi’an Jiaotong University, Xi’an, 710000, People’s Republic of China; 5Department of Hepatobiliary Surgery and Liver Transplantation, The Second Affiliated Hospital of Xi’an Jiaotong University, Xi’an, 710000, People’s Republic of ChinaCorrespondence: Chang Liu, Department of Hepatobiliary Surgery and Liver Transplantation, The Second Affiliated Hospital of Xi’an Jiaotong University, Xi’an, 710000, People’s Republic of China, Tel +86-29-87678231, Fax +86-29-87678599, Email [email protected] Jingyao Zhang, Department of SICU, The First Affiliated Hospital of Xi’an Jiaotong University, Xi’an, 710061, People’s Republic of China, Tel +86-29-85323900, Fax +86-29-85324642, Email [email protected]: This study aimed to establish a nomogram for early and accurate identification of infected pancreatic necrosis (IPN) among patients with acute necrotizing pancreatitis (ANP) by integrating clinical data and radiomic information from contrast-enhanced computed tomography (CECT).Patients and Methods: This retrospective single-center study included 203 ANP patients who underwent CECT. Patients were divided into training (n=142) and test set (n=61). Radiomic features were extracted from CECT images using PyRadiomics. Three machine learning classifiers were employed to construct a radiomic signature. Clinical factors were identified through regression analysis. A combined nomogram was developed using multivariate logistic regression. ROC and calibration curves were plotted to assess the efficacy of the model. Decision curve analysis (DCA) was applied to identify the clinical value and utility.Results: In the training and test set, 56 (39.43%) and 23 (37.70%) patients developed into IPN, respectively. The optimal Rad score was achieved by the LightGBM classifier. APACHE II and MCTSI scores were independent predictors of IPN. The combined clinical-radiomic nomogram achieved the best predictive efficacy, with an AUC of 0.877 in the training set and 0.829 in the test set. The calibration curve proved good accordance, and the decision curve demonstrated great clinical utility.Conclusion: The clinical-radiomic combined nomogram performed well in predicting IPN in patients with ANP. It could potentially serve as a quantitative, non-invasive tool for early IPN prediction in patients with ANP.Keywords: pancreatitis, infected pancreatic necrosis, contrast-enhanced computed tomography, radiomic, machine learning, nomogram