卷积神经网络在医学图像分割中的研究进展

卷积神经网络（CNN）是目前计算机视觉和模式识别中效果最为突出的算法。CNN拥有强大的空间识别能力，可以从图像中提取高阶的空间特征，同时通过共用卷积核的方式大幅减少参数量，从而在提升网络性能的同时保持总参数量在一个合理的、可运算的范畴。部分采用无监督学习的CNN算法可以在没有先验知识的条件下实现一定程度的图像语义分割，大幅减少人工读图的负担。本研究就CNN在医学图像分割中的研究进展和使用CNN时的具体技巧及其效果进行综述。以使用CNN为核心的深度学习工具解决医学图像分割的课题为中心，展示了CNN在有监督学习、半监督学习及无监督学习中的巨大潜力，分析比较了现有方案的优点与不足，探讨了未来CNN在医学图像领域的前进方向。

Progress on convolutional neural network in medical image segmentation

Convolutional neural network (CNN) is regarded as the state-of-the-art algorithm in computer vision and pattern recognition. CNN which is excellent for spatial recognition can extract hierarchical features from images. The number of parameters is greatly reduced by sharing kernels, thus improving the network performance and keeping the number of total parameters in a reasonable and computable range. To some extent, some CNN algorithms based on unsupervised learning can perform semantic segmentation without prior knowledge, releasing the burden of manual works. Herein the research progresses of CNN in medical image segmentation as well as the techniques and outcomes of CNN are reviewed. Focusing on the topic of using deep learning with CNN as core to realize medical image segmentation, the great potentials of CNN in supervised learning, semi-supervised learning and unsupervised learning are introduced. By analyzing and comparing advantages and shortages of existing methods, the prospects of CNN in medical imaging are discussed.