基于gpu加速的高斯混合模型的脑mri分割骆斌1骆仁波21湖北.docVIP

基于GPU加速的高斯混合模型的脑MRI分割 骆斌1 骆仁波2△ （1湖北省云梦县人民医院CTMRI室，湖北云梦 432500；2△华南理工大学 自动化科学与工程学院，广州510640） [摘 要] 目的 提高基于EM的高斯混合模型算法对数据量较大的3维脑MR图像分割时的效率，达到实时处理的要求。方法将基于EM的高斯混合模型与GPU并行运算相结合，并运用到了脑MR图像分割处理中。在分割过程中计算复杂高斯混合模型时，不是直接在CPU上做运算，而是移植到GPU上并行处理，加快图像分割处理速度。结果 实验表明基于GPU并行处理的高斯混合模型对脑MR图像分割速度比在CPU上要快4~8倍，并且分割处理数据量的越大，加速效果越明显。结论 擅长并行处理的GPU技术与高斯混合模型图像分割算法相结合，能加快医学人脑MR图像分割速度，满足医务人员临床诊断时实时性要求。 [关键词] GPU；高斯混合模型；核磁共振成像；图像分割 [中图分类号] TP391 [文献标识码] A MRI Segmentation Based on Gaussian Mixture Model accelerated by GPU Luo Bin1 Luo Renbo2 (1Department of CTMRI, Peoples Hospital of Yunmeng county，Yunmeng，432500，China; 2College of Automation Science and Engineering South China University of Technology, Guangzhou 510640,China) Abstract] Objective To improve the segmentation efficiency of Gauss Mixture Model based on EM when dealing with 3-dimensional brain Magnetic Resonance (MR) images,which with large amount of data, and meet the requirements of real-time processing. Methods The Gaussian mixture model based on EM and the GPU parallel computing method are combined and applied to the brain MR image segmentation processing. When Calculating the complex Gaussian mixture model during the segmentation process, not do operations directly on the CPU, but transplanted it to GPU and processed it by parallel mode, to accelerate the processing speed of image segmentation. Results Experimental results show that Gaussian mixture model based on parallel processing in GPU for brain MR image segmentation faster than sequential processing in CPU, and the speedup ratio reaches to 4~8 times. Whats more, the larger amount of data to be processed, the more obvious the effect is. Conclusion The combination of GPU parallel processing technology and the Gaussian mixture model image segmentation algorithm could accelerate the speed of brain MR image segmentation, meeting the medical staffs real-time requirements in the clinical diagnosis. [key words] GPU; Gaussian Mixture Model; Magnetic Resonance Ima