Objectives: To evaluate the efficacy of the radiomics model based on preoperative B-mode ultrasound (BMUS) and shear wave elastography (SWE) for predicting the occurrence of clinically relevant-postoperative pancreatic fistula (CR-POPF).
Methods: Patients who were scheduled to undergo pancreatectomy were prospectively enrolled and received ultrasound assessment within one week before surgery. The risk factors of POPF (grades B and grades C) were analyzed. Preoperative BMUS images, SWE values of pancreatic lesions and surrounding parenchyma were used to build preoperative prediction radiomics models. Radiomic signatures were extracted and constructed using a minimal Redundancy Maximal Relevance (mRMR) algorithm and an L1 penalized logistic regression. A combined model was built using multivariate regression which incorporated radiomics signatures and clinical data.
Results: From January 2020 to November 2021, a total of 147 patients (85 distal pancreatectomies and 62 pancreaticoduodenectomies) were enrolled. During the three-week follow-up after pancreatectomy, the incidence rates of grade B/C POPF were 28.6% (42/147). Radiomic signatures constructed from BMUS of pancreas parenchymal regions (panRS) achieved an area under the receiver operating characteristic curve (AUC) of 0.75, accuracy of 68.7%, sensitivity of 85.7 %, and specificity of 61.9 % in preoperative noninvasive prediction of CR-POPF. The AUC of the radiomics model increased to 0.81 when panRS was used for the prediction of CR-POPF after pancreaticoduodenectomy.
Conclusions: Radiomics model based on ultrasound images was potentially useful for predicting CR-POPF. Patients with high-risk factors should be closely monitored when postoperation.
Keywords: Postoperative pancreatic fistula; pancreatectomy; prediction; shear wave elastography; ultrasound radiomics.