基于深度学习的多模态医学图像融合方法研究进展

医学图像融合方法可以将有用的信息整合到一张图像上,提高单张图像的信息量。对多模态医学图像进行融合时,如何对图像进行有效的变换,提取到不同图像中独有的特征,并施以适当的融合规则是医学图像融合领域研究的重点。近年随着深度学习的快速发展,深度学习被广泛应用于医学图像领域,代替传统方法中的一些人工操作,并在图像表示、图像特征提取以及融合规则的选择方面显示出独特优势。本文针对基于深度学习的医学图像融合进展予以探讨,介绍卷积神经网络、卷积稀疏表示、深度自编码和深度信念网络这些常用于医学图像融合的框架,对一些应用于融合过程不同步骤的深度学习方法进行分析和总结,最后,分析当前基于深度学习的融合方法的不足并展望了未来的研究方向。

Advances in multimodal medical image fusion method based on deep learning

The medical image fusion method can integrate useful information into a single image to increase the amount of information in a single image.When multimodal medical images are fused, how to effectively transform the images, extract the special features in different images, and apply appropriate fusion rules becomes the focus of research on medical image fusion. In recent years, with the rapid development of deep learning, deep learning has been widely used in medical images by replacing some manual operations in traditional methods and has shown unique advantages in image representation, image feature extraction and the selection of fusion rules. Herein the medical image fusion based on deep learning is discussed. Several deep learning methods that are commonly used in the framework of medical image fusion, such as convolutional neural network, convolutional sparse representation, deep self-encoding and deep belief network are introduced. Some deep learning methods that are applied to different steps of the fusion process are also summarized. Finally, the disadvantages of recent researches on the image fusion based on deep learning are analyzed, and the research direction in future is forecasted.