Abstract:Objective To explore the method of deep convolutional neural network (CNN) combined with transfer learning (TL) to assist in the identification and diagnosis of brain cancer. Methods The basic principle of deep CNN and TL were introduced, and the model of deep CNN combined with TL was established. Brain CT images in our hospital were classified employing different models including FCNN, CNN, AlexNet, VGGNet, GoogLeNet, AlexNet-TL, VGGNet-TL and GoogLeNet-TL, and then the performance of each methods were evaluated based on sensitivity, specificity and accuracy. Results The pattern-recognition-accuracy-rate of craniocerebral cancer based on the model of FCNN, CNN, AlexNet, VGGNet and GoogleNet of with random initialization network parameters were 70.2%, 76.5%, 82.7%, 80.9% and 82.5%, respectively, while the pattern-recognition-accuracy-rate of craniocerebral cancer based on the model of AlexNet-TL, VGGNet-TL, and GoogleNet-TL with optimized model parameters after deep CNN pre-training using open source mega data set ImageNet were 86.9%, 90.2%, and 93.4%, respectively. Conclusion Intelligent identification and classification of craniocerebral cancer can be achieved using the model of deep CNN combined with TL, which may alleviate the working intensity of the medical staff. The method based on deep CNN combined with TL for the assistant diagnosis of craniocerebral cancer is effective and feasible.
蒋佳旺,陈艳,王佳庆. 卷积神经网络与迁移学习的颅脑癌症识别方法的研究[J]. 中国医疗设备, 2020, 35(9): 70-70.
JIANG Jiawang, CHEN Yan, WANG Jiaqing. Research on Recognition Methods of Brain Cancer Based on Convolutional Neural Network
and Migration Learning. China Medical Devices, 2020, 35(9): 70-70.