基于混合神经网络的脑部MRI图像语义分割算法

针对传统医学图像对缺乏标注的数据进行自动分割时存在分割精度不高、边缘模糊等问题，提出了一种利用混合神经网络对脑部核磁共振成像(magnetic resonance imaging，MRI)的图像进行语义分割的算法。利用仿射网络对脑部MRI图像进行线性几何变换，基于卷积神经网络进行3D医学图像仿射变换，加入稠密模块减轻梯度消失和加强特征传递问题; 通过空间转换网络对脑部MRI进行空间转换，基于图谱的分割法获得脑部图像的分割结果。采用MICCAI的公共数据集BraTs2019进行实验验证，结果表明，算法可由脑部肿瘤MRI图像获得较好的分割精度和分割效率，为脑部MRI图像语义分割的研究提供一种新的实验方案。

Semantic segmentation algorithm of brain MRI image based on hybrid neural network

Aiming at the problems of low segmentation accuracy and blurred edges in traditional medical image automatic segmentation on label-lacking data, this paper proposes a brain MRI image semantic segmentation algorithm based on hybrid neural network. This algorithm contains two-stage tasks. In the first stage, an affine network is used to perform linear geometric transformation on brain MRI images. The affine network is used to perform linear geometric transformation on brain MRI images, and the convolution neural network is used to perform affine transformation on 3D medical images. Dense modules are added to reduce the gradient disappearance and enhance the feature transfer. In the second stage, the brain MRI is spatially transformed through the Spatial Transformer Network. Finally, the segmentation result of the brain image is obtained by the segmentation method based on the atlas. The MICCAI''s public data set BraTs2019 is used for experimental verification. The results show that this algorithm has obtained good accuracy and efficiency for brain tumor segmentation from MRI images. The algorithm proposed in this paper provides a new way for accurate segmentation of medical images. Keywords: semantic segmentation; deep learning; medical image; atlas segmentation