摘要
针对2维超声心动图像噪声大且难于分割的特点,提出基于Centripetal Catmull-Rom曲线的多尺度活动形状模型的左心室分割方法。该算法在金字塔上层提取2维轮廓法线向量特征并采用马氏距离寻找新的特征点位置;在图像金字塔底层快速提取特征点周围的Log-Gabor特征并使用Gentle Ada Boost训练分类器,选择置信度水平最高的点作为新的特征点位置。实验证明,这种方法较之传统活动形状模型分割更为精确且分割结果交互方便,有利于对结果的再编辑。
In order to treat the problem that left ventricle is hard to segment in two-dimensional transesophageal echocardiography because of the speckle noise, a multi-scale active shape model based on centripetal Catmull-Rom spline was proposed. In the upper pyramid levels ,2D profiles were extracted for each landmark and Mahalanobis distance was used searching new position for each landmark. In the bottom level, fast Log-Gabor feature was extracted and trained using Gentle AdaBoost. The point with highest confidence-level was the new position for each landmark in the fitting process. Experiments proved that this method is more accurate than traditional Active Shape Model and the segmentation result is interactive and easy to edit.
出处
《四川大学学报(工程科学版)》
EI
CAS
CSCD
北大核心
2016年第5期84-90,共7页
Journal of Sichuan University (Engineering Science Edition)
基金
四川省科技创新苗子工程资助项目(2015060)
中科院西部之光人才培养计划资助项目