Objective:To optimize contrast-enhanced computed tomography(CECT)image quality using Non-Local Means(NLM)denoising technology and construct a mortality risk prediction model for severe acute pancreatitis(SAP)patients based on his-togram parameters.Methods:Retrospective analysis of clinical,laboratory,and CECT data of 74 patients diagnosed with SAP.NLM de-noising was applied to arterial and portal venous phase images at different coefficient levels.Three-dimensional volumes of interest were manually delineated in pancreatic parenchyma on original and denoised images,followed by histogram parameter extraction.After univa-riate Logistic regression feature selection,binary Logistic regression models were built to predict SAP mortality risk and compared with clinical models and conventional CT models.Results:A total of 74 patients with SAP,of whom 27(36.5%)died.After applying NLM-based noise reduction to arterial phase and portal venous phase CECT images,8 histogram prediction models were constructed under dif-ferent noise coefficient levels.There was no significant difference in predictive performance among these histogram models(P＞0.05).Compared to conventional CT models and clinical models,the histogram models demonstrated higher accuracy in predicting mortality risk(P=0.009,P=0.048).The predictive performance of the clinical model and the conventional CT model was similar(P=0.699).Conclusion:CECT histogram parameters show significant clinical value in predicting mortality risk in SAP patients,maintaining stable performance under different noise coefficient levels after NLM denoising.Their noise robustness and quantitative a-nalysis advantages compensate for the limitations of conventional CT.