Performance of CT Radiomics Analysis for Predicting Histopathological Features of Pancreatic Ductal Adenocarcinoma

Abstract

Purpose: To explore the performance of radiomics on contrast-enhanced CT for the preoperative prediction of histopathological features in pancreatic ductal adenocarcinoma (PDAC). Methods: This retrospective study included patients with surgically-resected PDAC and available preoperative contrast-enhanced CT imaging before any treatments. Histopathological analysis of resection specimens was used as reference standard for the diagnosis of lymph node metastasis, tumor grade, margins, perineural, and vascular invasion. Radiomics analysis was performed using a dedicated research software (LIFEx, version 5.10) on CT images acquired during the arterial and portal venous phases. Whole tumor segmentation was performed by one radiologist by drawing a region of interest in each CT slice within the lesion margins. Receiver operating curves, areas under the ROC (AUROC) and 95% confidence intervals (95%CI) were calculated to assess the accuracy of radiomics features for the prediction of histopathological features. Results: The study population included 49 patients (23 males, mean age 67.4 years) with PDCA (mean size 3.7 cm). On arterial phase, multiple radiomics parameters were significantly associated with presence of lymph node metastasis. Particularly, NGLDM_Coarseness demonstrated the highest diagnostic performance for the prediction of lymph node metastasis with an AUROC of 0.747 (95%CI 0.572-0.922, p = 0.018). On portal venous phase, multiple radiomics parameters were significantly associated with tumor grade, and CONVENTIONAL_HUstd showed highest diagnostic performance for the prediction of poorly differentiated (G3) PDAC with an AUROC of 0.865 (95%CI 0.712-1.000, p = 0.001). Conclusions: Radiomics analysis have a fair-to-good performance for the preoperative prediction of lymph node metastasis and tumor grade in PDAC.