A Preliminary Study on Mesothelin Expression Levels in Pancreatic Cancer Using a CT Radiomics Model

Abstract

Purpose To explore the feasibility of CT radiomics model in identifying the expression level of pancreatic mesothelin. Methods A retrospective analysis of 37 confirmed pancreatic cancer cases was conducted via surgical pathology. These cases had well-preserved tissue blocks and underwent upper abdominal CT scans within two weeks of surgery. Images, centered on the tumor's maximum diameter with one slice above and below the lesion, were selected for each case. Using 3D Slicer software, Regions of Interest (ROIs) were defined, and Pyradiomics extracted radiomics features. The dataset was categorized into positive and negative groups based on mesothelin immunohistochemical expression levels. Random division into training and testing sets ensued. Initial feature selection reduced radiomics dimensions, followed by secondary selection using (Least Absolute Shrinkage and Selection Operator) LASSO regression, resulting in a radiomics score model. Diagnostic performance was assessed in both sets using Receiver Operating Characteristic (ROC) analysis, precision via Calibration Curve (CC) analysis, and clinical benefit through Decision Curve Analysis (DCA). Results A total of 1218 radiomics features were extracted from 111 slices of CT scans in pancreatic cancer patients. The constructed radiomics model, after a series of dimensionality reduction and selection methods, achieved an Area Under the ROC Curve (AUC) of 0.84, sensitivity of 80.00%, and specificity of 75.68% in the training set. In the testing set, the AUC was 0.75, sensitivity was 58.82%, and specificity was 88.24%. The Calibration Curves (CC) in both the training and testing sets indicate a strong fit, while the Decision Curve Analysis (DCA) shows good clinical benefit. Conclusion A CT-based radiomics model can be used to evaluate pancreatic mesothelin expression levels, providing a reference for early prediction and differential diagnosis of pancreatic cancer using imaging techniques.