基于CUDA的医学图像处理算法平台的设计与初步实现

The Design and Realization of Medical Image Processing Algorithm Platform Based-on CUDA

目的:本文介绍了所开发的医学图像处理算法平台(cudaGIL),设计平台主要是为医学图像处理算法的开发和测试提供一个简洁的框架。方法:该平台封装了cudpp,cufft,thrust等第三方库,并提供简洁的算法接口,使得并行算法能在该平台上高效执行;采用了优化的迭代器模式和数据分页方法,用以降低数据索引的时间消耗;组件管理模式被用于该平台中,用户可以新建组件扩展自定义算法;使用建立在OpenGL与CUDA基础上的异步操作实时显示图像。结果:通过与已有平台的比较,结果显示该平台在算法效率和显示速度上有了较大的提升。结论:本文设计的医学图像处理算法平台可作为医学图像算法的开发工具。

Objective： In this paper we propose an image processing algorithm platform using GPGPU technique,which aims to provide a simple framework for medical image processing algorithm development and testing.Methods： This platform encapsulates cudpp,cufft,thrust and other third-party libraries,and provides simple algorithm interface.With the secure memory management mechanism,the parallel algorithm could be executed on the platform efficiently.Optimized iterator pattern and data paging management are used to decrease the data index time cost.Component management model is introduced in this paper,users can extend a new custom component algorithm and avoid compile the platform again.Images could be displayed in real-time with the asynchronous operation between OpenGL and CUDA.Results： Compared with other existed image processing platforms,our platform＇s data indexing speed and visualization performance are faster.Conclusions： The platform can be used as a development tool for medical image processing.