论文部分内容阅读
从超声图像序列中精确地分割出目标区域对于提高基于超声图像引导的计算机辅助治疗的效果具有重要的意义.活动轮廓模型被广泛地应用于医学图像的分割,但是超声图像固有的低信噪比、灰度分布不均匀等特点导致活动轮廓容易搜索到错误的区域.为此,学者们提出了不同的目标特征或形状先验知识来约束活动轮廓的演化.目前,大多数目标图像特征的工作仅限于提取目标区域信息的特征,目标形状先验也需要通过对大量训练样本中的目标形状进行学习,而且训练样本中形状是否能足够描述分割目标的形状信息也是值得怀疑的.本文提出了一种基于稀疏特征竞争和形状相似性的超声图像序列分割方法.该方法首先利用稀疏表达理论分别构建目标和背景的特征字典,并根据目标和背景特征字典的重构误差构建一种基于稀疏特征竞争的目标搜索策略,从而提高活动轮廓搜索真实边缘的鲁棒性.然后,对图像序列中目标形状的相似性进行建模,并证明了目标形状的变化符合低秩属性,该属性可看作是对图像序列中目标形状的先验知识进行无监督学习.为了验证该方法的性能,本文采用临床的超声图像序列作为训练和测试集合,并与其他3种典型的分割方法在同一测试集合下进行比较.实验结果显示,针对超声图像中出现的边缘模糊、缺失等缺点,该方法提供了更准确和鲁棒的分割结果,从而提高了计算机辅助治疗的效率和效果.
Accurate segmentation of target regions from ultrasound image sequences is of great importance to improve the effectiveness of computer-aided therapy based on ultrasound image guidance.Active contour models are widely used in medical image segmentation, but the inherent low SNR of ultrasound images , Uneven grayscale distribution and other characteristics lead to the active contour is easy to search for the wrong region.Therefore, scholars proposed different target features or shape prior knowledge to constrain the evolution of the active contour.At present, most of the target image features It is also doubtful to extract the feature of the information of the target region only if the target shape prior knowledge also needs to learn the target shape from a large number of training samples and whether the shape of the training sample can adequately describe the shape of the segmentation target.In this paper, Based on sparse feature competition and shape similarity, this method first constructs the feature dictionary of target and background using sparse representation theory and constructs a sparse feature based on the reconstructed error of object and background feature dictionary The goal of search strategy, thereby enhancing activities The robustness of the real edge of the contour search is studied.Then, the similarity of the target shape in the image sequence is modeled and it is proved that the change of the target shape conforms to the low-rank attribute, which can be regarded as the first of the target shape in the image sequence In order to verify the performance of this method, the clinical ultrasound image sequence is used as a training and testing set, and compared with the other three typical segmentation methods under the same test set.Experimental results show that, This method provides more accurate and robust segmentation results, thus improving the efficiency and effectiveness of computer-aided therapy.