A Preliminary Study Of CT Radiomics In Early Dignosis Of Pancreatic Cancer

Part 1 Application of CT-based radiomics for differentiating mass-forming chronic pancreatitis from pancreatic ductal adenocarcinoma in patients with chronic pancreatitisObjective: To develop and validate a CT nomogram and a CT-based radiomics nomogram to differentiate mass-forming chronic pancreatitis（MFCP）from pancreatic ductal adenocarcinoma（PDAC）in patients with chronic pancreatitis（CP）.Methods: In this retrospective study,the data of 138 patients with pathologically diagnosed MFCP or PDAC treated at our institution were retrospectively analyzed.Two radiologists analyzed the original cross-sectional CT images based on predefined criteria.Image segmentation,feature extraction,and feature reduction and selection were used to obtain the rad-score.The CT model and radiomics model were developed using a training set consisting of 44 patients with MFCP and 59 patients with PDAC who were diagnosed between February 2011 and April 2018.The models were validated in validation set consisting of 23 patients with MFCP and 12 patients with PDAC who were diagnosed between May 2018 and February 2021.The logistic regression models were adopted to establish respective nomogram.Nomogram performance was determined by its discriminative ability and clinical utility.Results: The mean age of patients was 53.7 years,75.4% were male.The multivariable logistic regression analysis showed that cystic degeneration,a duct-to-parenchyma ratio ≥ 0.34,the duct-penetrating sign,pancreatic portal hypertension and arterial CT attenuation were the independent predictors of differentiating between the two types of masses.The above independent predictors were selected for the logistic regression CT model.The CT model = 3.65-2.59×（cystic degeneration）+1.26×（duct-to-parenchyma ratio≥0.34）-1.40×（duct-penetrating sign）+1.36×（pancreatic portal hypertension）-0.05×arterial CT attenuation.The radiomics model = 3.14-3.62×（cystic degeneration）+1.04×（duct-to-parenchyma ratio≥0.34）-1.01×（duct-penetrating sign）+1.21×（Pancreatic portal hypertension）-0.04×arterial CT attenuation +3.54 radiomics score.The CT model-based nomogram showed good differentiation between the two entities in the training cohort（area under the curve [AUC],0.87）and in the validation cohort（AUC,0.94）.The radiomics model-based nomogram also showed good differentiation between the two entities in the training cohort（AUC,0.91）and in the validation cohort（AUC,0.93）.Decision curve analysis（DCA）showed that patients could benefit from the two nomograms.Conclusions: The two nomograms reasonably accurately differentiated MFCP from PDAC in patients with CP and hold potential for refining the management of pancreatic masses in CP patients.Part 2 Application of CT radiomics features in screening small pancreatic cancerObjective To develop a nonenhanced CT-based adjunctive diagnostic tool for screening small pancreatic cancer（maximum tumor diameter ≤2cm）to facilitate early diagnosis of pancreatic cancer.Methods In this retrospective study,206 patients with small pancreatic carcinoma confirmed by surgical pathology and 268 normal controls without known pancreatic disease were collected consecutively,and assigned to the training set and validation set in chronological order.144 patients with small pancreatic cancer and188 normal controls admitted between 2014 and 2019 were assigned to the training set.62 patients with small pancreatic cancer and 80 normal controls admitted between 2020 and 2021 were assigned to the validation set.The 3D volume of the pancreas was automatically segmented from the preoperative CT scans by nn U-Net,a self-configuring method for deep learning-based biomedical image segmentation.Radiomics features were extracted subsequently.Variance analysis,Spearman correlation analysis and receiver operating characteristic curve（ROC）analysis were applied to select features.The extreme gradient boosting classifier（XGBoost）for binary classification of small pancreatic cancer versus normal pancreas was developed using a training set and validated in the validation set.The performance of the XGBoost classifier was determined by its discriminative ability and clinical usefulness.Result Mean tumor size was 1.69 ± 0.34（SD）cm.The Dice coefficient of the automatic segmentation model is 0.85.The radiomics features included in XGBoost classifier were firstorder Skewness,firstorder_Median,firstorder_Mean,diagnostics_Image-original_Maximum,diagnostics_Image-original_Minimum,firstorder_Interquartile Range,firstorder₉0Percentile,firstorder₁0Percentile.The area under curve（AUC）,sensitivity,specificity,positive predictive value（PPV）and negative predictive value（NPV）of the XGBoost classifier in the training set were0.99,0.92,0.97,0.91 and 0.98,respectively.The AUC,sensitivity,specificity,PPV,NPV of the XGBoost classifier in the validation set were 0.99,0.94,0.94,0.93 and0.97,respectively.Decision curve analysis（DCA）showed that patients could benefit from the XGBoost classifier.Conclusion: Radiomics features extracted from whole pancreas of nonenhanced CT scan of the abdomen can be used to differentiate small pancreatic cancer from normal pancreas.We developed a CT-based XGBoost classifier which is useful in screening for small pancreatic carcinoma.