GPU加速的交互式医学CT图像区域分割

GPU-accelerated interactive medical CT image segmentation

目的为了解决交互式医学CT图像区域分割问题,本文提出了基于证据推理规则的区域生长算法(ERRG)。方法算法综合考虑了医学图像的灰度直方图,Gabor特征和灰度共生矩阵能量3个重要特征,采用Bhattacharyya系数度量相邻像素的相似程度,用效用函数将度量系数合并。针对算法计算效率较低问题,对算法进行并行化,采用GPU进行加速处理。结果本文算法与基于Random-Walk图像分割算法针对医学CT胃部图像,进行对比实验,表明使用本文算法,真阳性目标像素数占目标区域所有正确像素数的比例(TPF)显著提高,背景像素错误地分割为目标像素的数目占背景正确像素数的比例(FPF)显著降低;通过GPU加速后,算法执行效率显著提高,加速比达到12。结论本文算法减少了医学CT图像过分割现象,采用GPU加速后能够实现实时交互式医学CT图像分割。

Objective This paper proposes a novel evidential reasoning based region growing （ERRG） method to solve the segmentation problem of an interactive medical CT image. Method ERRG considers some important features of medical ira- ages, such as gray histogram, Gabor, and gray level co-occurrence matrix. The Bhattacharyya coefficient is used to meas- ure the similarity between the adjacent pixels and the utility function and to merge the metric coefficients. However, given the low efficiency of ERRG, a parallel region segmentation algorithm for interactive medical images is mapped to GPU to ac- celerate the algorithm. Result The true-positive fraction （TPF） can significantly increase, false-positive fraction （FPF） can significantly decrease, and the speedup is 12. Conclusion Real-time interactive medical image segmentation can be a- chieved using GPU-accelerated.