基于优选集成ConvNet的脑癌图像分割方法  

An optimization integration ConvNet method for brain cancer image segmentation

作　　者：韩兵 王鹏 周毅 HAN Bing;WANG Peng;ZHOU Yi(Center of Information,Beijing Shijitan Hospital Affiliated to Capital Medical University,Beijing 100038,China)

机构地区：[1]首都医科大学附属北京世纪坛医院信息中心,北京100038

出　　处：《安徽大学学报（自然科学版）》2022年第6期99-108,共10页Journal of Anhui University(Natural Science Edition)

基　　金：国家自然科学基金资助项目(61571361)。

摘　　要：基于对脑癌患者的诊断与治疗的重要性,提出一种优选集成卷积神经网络方法以对多模态的脑肿瘤图像进行有效而精准的分割.具体来说,根据所使用的2个子卷积神经网络的不同特点及其表现,将它们分别用于不同脑肿瘤模态数据的图像分割处理.此外,为了进一步提升分割的有效性,对原始网络进行了一些结构与超参数方面的优化.与脑肿瘤分割比赛中的优秀结果以及其他优秀的脑肿瘤图像分割方法进行比较,验证结果显示,笔者提出的优选集成卷积神经网络方法进行了脑肿瘤的高度精确分割.Based on the importance of diagnosis and treatment of patients with brain cancer, this paper proposed an optimization integrated convolutional neural network method for effective and accurate segmentation of multimodal brain tumor images. Specifically, according to the different characteristics and performance of the two convolutional neural networks used, they were respectively used for image segmentation of different brain tumor modal data. In addition, in order to further improve the effectiveness of segmentation, some structural and hyperparameter optimization of the original network was carried out. Compared with the excellent results in the brain tumor segmentation competition and other excellent brain tumor image segmentation methods, the verification results showed that the proposed optimal integrated convolutional neural network method performs highly accurate segmentation of brain tumor.

关 键 词：脑癌 脑肿瘤 优选集成 图像分割 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]