Abstract: Arterial Spin Labeling（ASL） MR images suffer from low SNR. In order to achieve adequate high image quality, it is necessary to execute experiment a number of times repeatedly. 3D Gaussian filter is widely in clinical use but the filter effect is not good enough. A new ASL image denoising algorithm based on Non-Local Means（NLM） denoising method is proposed. The new denoising method uses region similarity weight to decrease SNR and increases the accuracy of Cerebral Blood Flow（CBF） map. The results of CBF maps indicate that the CBF map obtained by the proposed method is more close to the real result than the Gaussian smoothing, which realizes the goal of obtaining accurate CBF map in the case of less repeated times.

Key words: Non-Local Means（NLM）, Gaussian filtering, Arterial Spin Labeling（ASL）, image denoising

摘要： 动脉自旋标记（ASL）MR图像信噪比低，需要重复采集多次以获得高质量的血流（CBF）图。临床中通常使用3D高斯滤波降低噪声但效果不佳。鉴于此，提出了新的基于非局域均值滤波（NLM）的ASL图像去噪方法，利用图像内部块相似度加权，降低噪声并提高血流图的计算精度。实验证明：与高斯滤波的结果比较，采用新方法得到的血流图和真实结果更接近，实现了在较少采集次数的情况下，得到精确的血流图像的目标。

关键词: 非局部均值方法（NLM）, 高斯滤波, 动脉自旋标记（ASL）, 图像去噪