基于SURF算法的HIFU治疗前后超声图像配准研究

高强度聚焦超声(HIFU)治疗过程中,医生通常将治疗前后的超声监控图像直接相减,得到治疗区域的图像来判断治疗效果。由于治疗后病人的体位可能会变化,导致治疗前后采集的图像发生偏移,通常需要先将HIFU治疗前后的超声监控图像进行配准。提出了一种基于加速鲁棒性算法(SURF)和超声监控图像散斑特征的快速图像配准方法。首先利用经典的SURF特征检测器分别提取治疗前和治疗后超声图像中的特征点,采用欧式距离匹配提取的特征向量,通过超声图像的固有属性散斑特征的分布规律优化配准参数并用RANSAC算法去除错误匹配点对,最后利用最小二乘法求出图像之间的映射关系完成配准。实验结果表明,该算法能自动搜索治疗前后超声监控图像中的对应特征点,比传统的人工筛选特征点效率更高,且利用散斑特征的分布规律优化后的SURF算法相比标准的SURF算法配准精度更高。

Study on ultrasonic image registration before and after HIFU treatment based on SURF algorithm

During high intensity focused ultrasound (HIFU) treatment, the monitoring image before and after the treatment were subtracted directly to obtain the image of the treated area and determine the therapeutic effect. Since the patient''s position may be changed after the treatment resulting in that the acquired images are shifted, the registration of ultrasonic monitoring image before and after HIFU treatment is necessary. In this paper, a fast ultrasonic monitoring image registration method based on speeded up robust features (SURF) algorithm and speckle features is proposed. First of all, the feature points are extracted from the monitoring image before and after the treatment with classical SURF feature detector. To optimize registration parameters, we make full use of the inherent properties of speckle features distribution of the ultrasound image. Then we use the Euclidean distance to match the feature vectors and eliminate erroneous matching points with the random sample consensus (RANSAC) algorithm. Finally, the relationship between the images is mapped to complete registration with the least squares method. The experimental results show that the algorithm can automatically search corresponding feature points from the ultrasonic monitor image before and after treatment which is more efficient than the traditional manual screening of the feature point and more accurate than the standard algorithms.