Application value of RefineNet convolutional neural network in accurate diagnosis of skull base fracture on CT

LIN Yuwen; HUANG Dongyun; LONG Lin

Abstract：Objective To explore the application value of RefineNet convolutional neural network in the accurate diagnosis of skull base fractures on CT.Methods A total of 90 patients who underwent head CT imaging in our hospital from June 2019 to June 2022 were selected.According to analysis,46 patients were found to have skull base fracture(skull base fracture group in this study),and 44 patients were normal brain(control group in this study).Patient characteristics were analyzed to define RefineNet structural parameters;The performance differences between Efficient Net,VGG16,Dense Net,Res net-50 and Xcep-tion and RefineNet were compared and analyzed.Meanwhile,the performance differences between five existing methods,includ-ing Mobile Net,Inception-v3,Dense NET-201,CUMED and SDL and RefineNet,were analyzed.RefineNet convolutional neu-ral network and manual testing recall,accuracy,and testing time were verified.Results The accuracy and AUC values of Re-fineNet were significantly higher than those of the five models and the five existing methods.The recall and accuracy of RefineNet convolutional neural network in fracture patients,skull base region,anterior cranial region,middle cranial region and posterior cranial region were significantly higher than those of manual test(P<0.05).The testing time required by RefineNet convolutional neural network was significantly less than that required by manual testing(P<0.05).Conclusion RefineNet convolutional neu-ral network has high accuracy and less time required for CT diagnosis of skull base fractures.