Predicting Postoperative Pancreatic Fistula after Pancreatoduodenectomy using Continuous-Time Random-Walk DWI and Virtual MR Elastography

Motivation: Clinically relevant postoperative pancreatic fistula (CR-POPF) is one of the most threatening complications after pancreatoduodenectomy, and there remains challenging to accurate predict CR-POPF preoperatively. Goal(s): To investigate the value of continuous-time random-walk (CTRW) and virtual MR elastography (vMRE) in predicting CR-POPF after pancreatoduodenectomy in pancreatic ductal adenocarcinoma. Approach: Independent quantitative parameter from CTRW and vMRE associated with CR-POPF were identified, and a combined prediction model were built. Results: CTRW_β (OR = 7223.62), CTRW_Dm (OR = 3117.57), and vMRE_μDiff (OR = 6.60) were independent preoperative predictors of CR-POPF (all p < 0.05). The combined model demonstrated satisfied performance with an AUC of 0.866. Impact: Non-Gaussian diffusion model CTRW and vMRE techniques exhibited the ability in predicting CR-POPF preoperatively, and their combination could improve the predictive performance. Our prediction model may help clinicians adopt appropriate prevention or mitigation strategies to reduce the probability of CR-POPF.