Radiomics-based histological grading of pancreatic ductal adenocarcinoma using F-FDG PET/CT: A two-center study.

OBJECTIVE

To explore the value of radiomics features derived from F-FDG PET/CT images in predicting the histological grade of pancreatic ductal adenocarcinoma (PDAC).

MATERIALS AND METHODS

A retrospective analysis was conducted using data from patients with suspected pancreatic cancer, who histologically confirmed as PDAC within 14 days after F-FDG PET/CT scan in one of two hospitals. Tumors were divided into high-grade (undifferentiated or poorly differentiated), and low-grade (moderately or well differentiated). Two researchers independently used uRP to perform layer-by-layer tumor segmentation in both PET and CT images of each patient, and extract features. Model performance was evaluated using 5-fold cross-validation on the entire multi-center cohort, with results averaged across all folds. The least absolute shrinkage and selection was used for feature selection, and support vector machine (SVM), random forest (RF), and logistic regression (LR) were employed to distinguish the grade of PDAC. The performance of the model was evaluated using the receiver operating characteristic curve.

RESULTS

This study comprised 111 patients (72 males and 39 females), comprising 52 patients with high-grade PDAC tumors and 59 patients with low-grade. A series of models were established by SVM, LR, and RF algorithms based on selected features. In the test set, the mean areas under the curve (AUCs) for PET image-based models using SVM, LR, and RF algorithms were 0.773, 0.772, and 0.760. For CT-based models, the mean AUCs were 0.764, 0.770, and 0.576. For PET/CT-based models, the mean AUCs were 0.840, 0.844, and 0.773.

CONCLUSION

Despite the lack of external validation, the PET/CT-derived radiomics model enables accurate preoperative histological grading of PDAC, offering a clinically actionable tool to neoadjuvant therapy stratification and further guide personalized medical decision-making.