Intelligent Assisted Diagnosis Model For Acute Pancreatitis Based On Deep Learning

Objectives: The aim of this study is to develop an intelligent assisted diagnosis model for acute pancreatitis based on Deep Learning(DL)algorithm.Materials and methods: This study included 190 patients with acute pancreatitis admitted to the Sichuan Emergency Center of Sichuan Provincial People’s Hospital from January 2020 to December 2021,and the first clinical and laboratory data within24 hours were collected,and the differences in baseline data between patients with acute mild pancreatitis and acute severe pancreatitis were analyzed by SPSS.In addition,abdominal CT images of patients with acute pancreatitis and healthy physical examination patients were collected.The lightweight deep learning network algorithm Mobile Net was used to establish an acute pancreatitis classifier model.The diagnostic effectiveness of the model was evaluated by the Area under curve(AUC).Furthermore,based on the above work,acute pancreatitis lesion segmentation model was constructed through U-Net network to assist the diagnosis of the severity of acute pancreatitis,and accuracy,loss,frequency-weighted accuracy and Mean Intersection over Union(MIOU)were used to evaluate the segmentation efficiency of the model.Result: Retrospective study showed that there were significant differences between acute mild pancreatitis patients and acute severe pancreatitis patients in inflammatory indexes,liver and kidney function,coagulation function and other aspects on admission(P < 0.05).The acute pancreatitis classifier model showed good diagnostic efficacy in the validation set [AUC 0.970(95%CI: 0.921-0.993)in the validation set for patients with healthy physical examination vs.patients with acute pancreatitis,P<0.001];[AUC 0.948(95%CI: 0.907-0.975)in the external validation set for patients with healthy physical examination vs.patients with acute pancreatitis),P < 0.001].Acute pancreatitis lesion segmentation model also showed high segmentation performance in the validation set.In the segmentation model of pancreas,peripancreatic inflammatory exudation,peripancreatic effusion and peripancreatic abscess necrosis,Their MIOU were 86.02(84.52,87.20),61.81(56.25,64.83)],57.73(49.90,68.23),66.36(55.08,72.12)].Conclusion: The intelligent auxiliary diagnosis model of acute pancreatitis based on deep learning has good diagnostic efficacy and can accurately distinguish the severity of acute pancreatitis.