计算机科学与探索 ›› 2018, Vol. 12 ›› Issue (4): 608-617.DOI: 10.3778/j.issn.1673-9418.1704042

• 人工智能与模式识别 • 上一篇    下一篇

结合卷积神经网络和模糊系统的脑肿瘤分割

师冬丽,李  锵+,关  欣   

  1. 天津大学 微电子学院,天津 300072
  • 出版日期:2018-04-01 发布日期:2018-04-04

Brain Tumor Image Segmentation Algorithm Based on Convolution Neural Network and Fuzzy Inference Syste

SHI Dongli, LI Qiang+, GUAN Xin   

  1. School of Microelectronics, Tianjin University, Tianjin 300072, China
  • Online:2018-04-01 Published:2018-04-04

摘要: 为了提高脑肿瘤分割的精确性和鲁棒性,提出一种结合卷积神经网络和模糊推理系统的全自动脑肿瘤MRI图像分割算法。首先,分别针对FLAIR和T2两种类型的单模态图像,构建适用于该类型图像的卷积神经网络。其次,针对FLAIR和T2图像,分别应用其对应的卷积神经网络模型进行预测,并将得到的预测概率通过非线性映射进行处理。最终,构建模糊推理系统,将FLAIR和T2图像经过非线性映射后的概率作为模糊推理系统的输入来判断该像素点是否属于肿瘤区域。实验结果表明,相比已有的脑肿瘤MRI图像分割算法,所提算法在分割精度上有了一定程度的提升。

关键词: 脑肿瘤, 图像分割, 卷积神经网络, 模糊推理系统

Abstract: In order to improve the accuracy and robustness of brain tumor image segmentation, this paper proposes an automatic segmentation algorithm of brain tumor MRI image based on convolution neural network and fuzzy inference system. Firstly, the convolution neural networks for the two types of single mode images of FLAIR and T2 are constructed. Then, the corresponding convolution neural network model is applied to each type of images to obtain the predicted probability which is processed by nonlinear mapping. Finally, the nonlinear mapped probability of the two kinds of images of FLAIR and T2 is used as the input of fuzzy inference system to determine whether the pixel belongs to the tumor area. Compared with existing brain tumor MRI image segmentation algorithm, the experimental results show that the proposed algorithm has an improvement in segmentation accuracy.

Key words: brain tumor, image segmentation, convolution neural network, fuzzy inference system