模态桥传递学习方法在医学图像分类中的应用

为减轻医疗领域中标签数据不足的问题。本文提出了一种基于转移学习的医疗图像分类的新方法。提出了一种桥梁数据库与目标数据库采用相同的医学成像采集方式的模式桥传输学习方法,通过学习从源到桥和从桥到目标的投影函数,可以减轻源(例如自然图像)和目标(例如X射线图像)之间的区域差异。实验结果表明,与各种传输学习方法相比,所提出的方法即使对于少量标记的目标医学图像也能够实现高分类性能。

To alleviate the problem of insufficient label data in medical field.In this paper,a new method of medical image classification based on transfer learning is proposed.A model bridge transmission learning method is proposed in which the bridge database and the target database adopt the same medical imaging acquisition mode.By learning the projection functions from the source to the bridge and from the bridge to the target,the regional differences between the source(such as the natural image)and the target(such as X ray images)can be reduced.The experimental results show that,compared with various transmission learning methods,the proposed method can achieve high classification performance even for a small number of labeled target medical images.