Prediction of synchronous distant metastasis of primary pancreatic ductal adenocarcinoma using the radiomics features derived from 18F-FDG PET/MR imaging

Abstract

Abstract Objective Despite the improved lesion detectability as the outcome of 18F-FDG PET/MR, small distant metastasis of pancreatic ductal adenocarcinoma (PDAC) often remains invisible. Our goal is to explore the potential of the joint radiomics analysis of PET and MRI imaging (PET-MRI) of primary tumors for predicting the risk of distant metastasis in patients with PDAC. Methods Nighty one PDAC patients with 18F-FDG PET and MRI imaging before the confirmation or exclusion of SDM were retrospectively investigated. Among them, 66 patients who received 18F-FDG PET/CT and multi-sequence MRI separately were included in the development of the radiomics model (development cohort), and 25 patients scanned with hybrid PET/MR were incorporated for independent verification (external test cohort). A radiomics signature was constructed using the selected PET-MRI radiomics features of primary PDAC tumors. Furthermore, a radiomics nomogram was developed by combining the radiomics signature and clinical indicators assisting in this way in the assessment of patients’ metastasis risk. Results In the development cohort, the radiomics nomogram had a better performance in predicting the risk of distant metastasis [area under the curve (AUC): 0.93, sensitivity:87.0%, specificity:85.0%] than this of the clinical model (AUC: 0.70, P < 0.001; sensitivity: 70%, specificity: 65%), as well as of this of the radiomics signature (AUC: 0.89, P > 0.05; sensitivity: 65%, specificity: 100%). For the external test, the radiomics nomogram yielded an AUC of 0.85, a sensitivity of 78.6%, and a specificity of 90.9%, which was comparable to the development (P = 0.34). Conclusions The preliminary results confirmed the potential of PET MRI-based radiomics analysis in the robust and effective prediction of the risk of SDM for preoperative PDAC patients. The in-depth analysis of the primary tumor may offer complementary information and provide hints for cancer staging.