生物力学模型约束下的图像分割算法研究

Biomechanical Model Constrained Image Segmentation Framework

引入基于生物力学约束的弹性可变形体模型,提出区别于传统Snake方法的内外能量构造方法,其中内能量取决于该弹性可变形体材料的物理特性,外能量是由图像等外部数据驱动,通过能量最小化过程达到平衡位置,即所要的分割结果。为了说明该算法的精确性和稳定性,分别在人工合成图上添加不同数量的噪声,并进行边界弱化。实验表明,该算法具有良好的分割效果。心脏核磁共振(MRI)图像分割的实验表明,该模型能有效地同时搜索到左心室的内外膜,解决了心外膜不易分割的难题。

A deformable model based on biomechanical constraints is introduced. The new framework has a form of internal and external energy term which is different from traditional Snake method. A biomechanical model formulation provides the representation and computation platform for the internal energy, and the gradient vector flow integrated with the region-based force forms the external energy. Through an energy minimization process, the external energy reaches its equilibrium,the expected segmentation result. Experiments on synthetic images with different amount of noise or weak edges show the accuracy and robustness of the algorithm. Experiments on real images of MR show the algorithm is effective in simultaneously finding the boundary of endocardium and epicardium,which solve the difficult problems in segmenting epicardium boundary of the heart.