多分辨超声心动图像分割模型

提出了一种新的基于水平集的多分辨空间超声心动图像分割模型，该模型在粗尺度上实现预分割，然后通过解传递方法将结果传递到细尺度上进行优化分割．在粗尺度上采用基于区域的高斯噪声模型分析图像，并和测地线模型相结合实现预分割，预分割结果表明了组合模型能自动和精确地提取边界．提出了基于数学形态学算子的尺度间快速解传递方法，该方法不需要进行插值运算，避免了常规方法效率低的问题．在细尺度上提出了一种局部活动轮廓优化模型，定义了新的基于局部亮度的目标函数．优化后的平均相似性从0．9862提高到0．9985．对左心室图像的分割实验证明了多分辨模型和常规方法相比具有更好的精确性和鲁棒性．

Multi-Resolution Segmentation Model for Echocardiographic Image

A novel level-set-based multi resolution segmentation method for echocardiographic images is presented. The algorithm begins with pre-segmentation at a low resolution level and transmits the result to optimally segmentalize on a higher resolution scale. At the low resolution, a region-based Gaussian noise model is applied to analyze the echocardiographic images and combined with a geodesic contour model for the pre-segmentation, while the combinative model is competent to accurately and automatically extract the boundary. A fast mathematical morphology-based approach is developed to transmit the pre-segmentation result where interpolation becomes unnecessary. At the high resolution, a local active contour optimization model and a new intensity-based objective function are proposed, and the average similarity measure increases from 0. 986 2 to 0. 998 5. Segmentation of left ventricle images shows the higher accuracy and robustness of the proposed multi-resolution segmentation method than the conventional ones.