基于卷积神经网络的皮肤癌良恶性预测  

Prediction of Benign and Malignant Skin Cancer based on Convolutional Neural Network

作　　者：董青青 银温社 易三莉[1] DONG Qing-qing;YIN Wen-she;YI San-li(School of Information Engineering and Automation,Kunming University of Science and Technology,Kunming Yunnan 650500,China)

机构地区：[1]昆明理工大学信息工程与自动化学院,云南昆明650500

出　　处：《通信技术》2018年第9期2081-2086,共6页Communications Technology

基　　金：国家自然科学基金项目(No.11265007)~~

摘　　要：皮肤癌是世界上最致命的癌症之一,早期诊断意义重大。为了改进皮肤癌的识别效果,在迁移学习模型CIFAR-10基础上,提出了一种改进的卷积神经网络模型。首先,在CIFAR-10增加一个卷积层和一个池化层;其次,通过反向传播的输出值与期望值调整网络参数;最后,不断调试网络,选择最佳模型作为识别网络。实验结果表明,改进的卷积神经网络训练集预测精度为91.92%,测试集预测精度为89.5%,为皮肤癌良恶性预测提供了一个检测手段,有望成为医生诊断皮肤癌的辅助诊断工具。Skin cancer is one of the most deadly cancers in the world,and its early diagnosis is of significant importance.In order to improve the recognition effect of skin cancer,based on the migration learning model CIFAR-10,a modified convolutional neural network model is proposed.Firstly,a convolutional layer and a pooling layer are added to CIFAR-10.Then,the network parameters are adjusted by the back-propagated output values and expected values.Finally,by continually debugging the network,the best model is selected as the identification network.The experimental results indicate that the prediction accuracy of the modified convolutional neural network training set is 91.92%,and the prediction accuracy of the test set 89.5%,and this provides a means for detecting benign and malignant skin cancer and is expected to become a diagnostic tool for doctors to diagnose skin cancer.

关 键 词：深度学习 黑色素瘤 图像分类 卷积神经网络 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]