代闯闯* **,栾海晶* **,杨雪莹* **,过晓冰***,牛北方* **,陆忠华* **.基于卷积神经网络的皮肤病诊断多二分类器研究[J].高技术通讯(中文),2022,32(10):1025~1035 |
基于卷积神经网络的皮肤病诊断多二分类器研究 |
Research on multi binary classifier for skin disease diagnosis based on convolutional neural network |
|
DOI:10.3772/j.issn.1002-0470.2022.10.004 |
中文关键词: 辅助医学诊断; 卷积神经网络(CNN);皮肤病;多二分类器 |
英文关键词: auxiliary medical diagnosis, convolutional neural network (CNN), dermatosis, multi-binary classier |
基金项目: |
作者 | 单位 | 代闯闯* ** | (*中国科学院计算机网络信息中心北京100190)
(**中国科学院大学北京 100049)
(***联想研究院北京 100094) | 栾海晶* ** | (*中国科学院计算机网络信息中心北京100190)
(**中国科学院大学北京 100049)
(***联想研究院北京 100094) | 杨雪莹* ** | (*中国科学院计算机网络信息中心北京100190)
(**中国科学院大学北京 100049)
(***联想研究院北京 100094) | 过晓冰*** | (*中国科学院计算机网络信息中心北京100190)
(**中国科学院大学北京 100049)
(***联想研究院北京 100094) | 牛北方* ** | (*中国科学院计算机网络信息中心北京100190)
(**中国科学院大学北京 100049)
(***联想研究院北京 100094) | 陆忠华* ** | (*中国科学院计算机网络信息中心北京100190)
(**中国科学院大学北京 100049)
(***联想研究院北京 100094) |
|
摘要点击次数: 1553 |
全文下载次数: 1057 |
中文摘要: |
近年来,随着深度学习技术的日益普及与发展,卷积神经网络(CNN)被广泛应用于辅助医学诊断,并在医学影像诊断领域取得了重要的研究成果。本研究基于皮肤病数据种类繁多、特征不显著等特点,引入多二分类的研究方法搭建了从医学影像到计算机辅助诊断的框架,解决了目前皮肤病难以区分的问题,并在常见的皮肤病分类识别问题中得到具体检验。首先,本研究以3类常见的皮肤病数据集(白癜风、痤疮和银屑病)为例,实现了图像数据的增强、分割、多二分类器的构建、图像块的分类、皮肤病的判别等完整的工作流程。其次,在分组数据交叉验证下,三分类判别准确率为0.8320,四分类判别的准确率达到0.9125。最后,为了获得更高的准确率,在随机森林方法结果不理想的情况下,本研究引入了多二分类器网络架构,准确率达到了0.9377。 |
英文摘要: |
In recent years, with the development of deep learning technology, the convolutional neural network (CNN) has been applied to assist medical diagnosis in some medical fields and has made great progress in the field of medical imaging diagnosis. In this study, based on the characteristics of skin disease data, such as a wide variety of features are not significant, a multi binary classification method is introduced to build a framework from medical images to computer-aided diagnosis, which overcomes the difficulty of distinguishing skin diseases, and has been tested in common skin disease classification and recognition. First of all, this study takes three kinds of common skin disease data sets (vitiligo, acne, psoriasis) as examples to realize the complete workflow of image data enhancement, segmentation, construction of multi binary classifier, classification of patches, skin disease discrimination. Secondly, under the cross validation of grouped data, the accuracy of three classification is 0.8320, and the accuracy of four classification is 0.9125. Finally, in order to obtain higher accuracy, in the case that the results of random forest method are not ideal, this study introduces a multi two classifier network architecture, and its accuracy reaches 0.9377. |
查看全文
查看/发表评论 下载PDF阅读器 |
关闭 |
|
|
|