基于PRW区域修正的交互式医学图像分割研究

Interactive Medical Image Segmentation Based on PRW Region Correction

由于医学图像的局部区域之间往往具有较高的相似度,使得其分割性能难以达到预期效果,为此提出基于PRW区域修正的交互式医学图像分割方法。首先根据交互式医学图像初始阶段的手工标注情况,利用RW算法对种子点进行划分,为简化传统RW计算,将划分过程转换为Dirichlet积分求解。然后利用PRW计算种子点的分布状况,根据目标图像的像素分布情况得到目标方程,并在其中引入先验能量,从而求解出种子点分类。最后基于邻近像素的相似性提出了邻域模型和区域修正策略,按照各区域的特征构建相应特征势能,采用连接检查调整分类错误的局部像素。通过仿真得到,方法的平均指标达到0.81,平均RDV指标达到0.17,平均分割时间为5.29s。结果表明所提方法能够在有限的手工标注前提下,完整、准确的实现医学图像分割,并且具有良好的分割速度。

Because of the high similarity between local regions of medical images,it is difficult to achieve the desired segmentation performance,Therefore,an interactive medical image segmentation method based on PRW region correction is proposed.Firstly,according to the manual annotation in the initial stage of interactive medical images,the RW algorithm was used to divide the seed points.In order to simplify the traditional calculation,the partition process was transformed into the solution of Dirichlet integral solution.Then PRW was used to calculate the distribution of seed points,and the target equation was obtained according to the pixel distribution of the target image,and a priori energy was introduced to solve the seed point classification.Finally,based on the similarity of adjacent pixels,the neighborhood model and region modification strategy were proposed.According to the characteristics of each region,the corresponding feature potential energy was constructed,and the connection check was used to adjust the local pixels with wrong classification.The simulation results show that the average DSC index is 0.81,the average RVD index is 0.17,and the average segmentation time is 5.29 s.The results show that the proposed method can achieve complete and accurate medical image segmentation under the premise of limited manual annotation,and has good segmentation speed.