摘要
基于深度学习的皮肤癌计算机辅助诊断技术在提高诊断准确率和效率方面取得了重大进展。以卷积神经网络为代表的深度学习技术已成为医疗影像分类识别中的主流技术。由于不同皮肤疾病图像特征的相似性,皮肤癌图像的分类问题属于图像的细粒度分类问题,使用传统卷积神经网络并不能达到预期的效果,因为它忽略了医疗影像数据集自身的特性。以ResNet50模型作为特征提取网络,设计Navigator、Teacher、Scrutinizer网络,构建面向细粒度皮肤疾病诊断的NTS-Net网络,并融合皮肤图像元数据特征(metadata)。对比于通用的卷积神经模型取得更好的效果,在ISIC国际皮肤癌图像数据集上达到90%以上的精度。
The computer-aided diagnosis technology based on deep learning has made significant progress in improving the accuracy and efficiency of diagnosis.The deep learning technology represented by convolutional neural network has become the mainstream technology in medical image classification and recognition.Due to the similarity of image features of different skin diseases,the classification of skin cancer images is a problem of fine-grained classification of images.The using of traditional convolutional neural networks cannot achieve the expected effect because it ignores the characteristics of the medical image data set itself.This paper used the ResNet50 model as a feature extraction network,designed Navigator,Teacher,and Scrutinizer networks,built an NTS-Net network for fine-grained skin disease diagnosis,and integrated skin image metadata features.Compared with the general convolutional neural model,the proposed method achieved better results,and achieved more than 90%accuracy on the ISIC international skin cancer image data set.
作者
蔡立志
章伟
陈敏刚
王乃琪
Cai Lizhi;Zhang Wei;Chen Mingang;Wang Naiqi(Shanghai Key Laboratory of Computer Software Testing&Evaluating,Shanghai Development Center of Computer Software Technology,Shanghai 201112,China;School of Information Science and Engineering,East China University of Science and Technology,Shanghai 200237,China)
出处
《计算机应用与软件》
北大核心
2023年第6期140-146,204,共8页
Computer Applications and Software
基金
上海市科学技术委员会科研计划项目(17411952800,18DZ2203700)。
关键词
计算机辅助诊断
卷积神经网络
细粒度分类
Computer aided diagnosis
Convolutional neural networks
Fine-grained classification
