# Radiomics analysis of dual-energy CT-derived iodine maps for predicting histopathologic grading of pancreatic ductal adenocarcinoma: a two-center study

## TL;DR

This study shows that using radiomics features from dual-energy CT iodine maps can help predict the severity of pancreatic cancer before surgery.

## Contribution

A novel radiomics-clinical model combining DECT-derived iodine map features with clinical factors for preoperative PDAC grading prediction.

## Key findings

- The radiomics-clinical model achieved AUCs of 0.873, 0.836, and 0.862 in training, testing, and external validation datasets.
- Combining radiomics features with BMI and CA125 levels improved overall predictive performance and reliability.

## Abstract

This study aimed to develop and validate a nomogram with radiomics features extracted from dual-energy computed tomography (DECT)-derived iodine maps for preoperatively predicting the histopathologic grading in pancreatic ductal adenocarcinoma (PDAC).

In this two-center retrospective analysis, 151 patients were enrolled (82 in the training set; 36 in the testing set, and 33 in the external validation set), all of whom underwent DECT imaging. The radiomics signature was developed using features extracted from DECT-derived portal venous phase (PVP) iodine maps. A clinical model was subsequently established based on significant clinical factors identified through multivariate analysis. The radiomics signature combined with clinically significant features was used to construct the final predictive model. Model performance was assessed through the area under the receiver operating characteristic curve (AUC) and decision curve analysis (DCA). The most predictive model was employed to construct the nomogram, with its calibration accuracy being assessed through calibration plot analysis.

The radiomics-clinical model, combining the radiomics signature, body mass index, and carbohydrate antigen 125 levels, showed strong predictive performance for predicting histopathologic grade in PDAC across the training, testing, and external validation datasets, with respective AUCs of 0.873, 0.836, and 0.862. The DCA and calibration curve demonstrated an enhanced overall benefit and demonstrated reliable consistency.

The radiomics–clinical model exhibited strong performance in preoperatively predicting the histopathologic grading in patients with PDAC.