基于粒子滤波的透射电镜肾小球基底膜分割

发布时间：2018-08-02 11:36

【摘要】：为了分割出厚度突然变窄、走向突然转变的基底膜,以辅助病理医生诊断遗传性肾小球疾病,提出了一种基于粒子滤波的透射电镜(TEM)肾小球基底膜分割方法。依据粒子滤波的原理,首先手动选择小段基底膜得到初始化的粒子;然后根据已估计状态的粒子的先验信息,利用动态模型产生下一个状态的粒子集;再从中选择满足约束条件的粒子;并利用似然函数选出权重较大的粒子;在连续状态的估计中,为了分割出厚度突然变窄、走向突然转变的基底膜,结合了回溯法和重采样的方法来更新粒子;随着多个连续状态的粒子估计,最终得到基底膜的分割结果。实验结果:选择了6组厚度突然变窄、走向突然转变的肾小球TEM图像进行基底膜分割,Dice相似系数范围为0.70~0.85。方法能够有效分割出厚度突然变窄、走向突然转变的基底膜,得到较为完整的基底膜。
[Abstract]:In order to segment the basement membrane whose thickness suddenly became narrower and which changed suddenly to assist pathologists in the diagnosis of hereditary glomerular diseases, a method of glomerular basement membrane segmentation based on particle filter for transmission electron microscope (TEM) was proposed. According to the principle of particle filter, the initialized particles are obtained by manually selecting the small segments of the substrate membrane, and then the particle set of the next state is generated by using the dynamic model according to the prior information of the particles in the estimated state. Then select the particles that meet the constraint conditions, and use the likelihood function to select the high-weight particles. In the estimation of the continuous state, in order to segment the basement membrane that the thickness suddenly narrows, the substrate changes suddenly. The backtracking method and the resampling method are combined to update the particles, and with the estimation of multiple continuous states, the segmentation results of the substrate membrane are obtained. Results: six groups of glomerular TEM images with suddenly narrow thickness and sudden change were selected to segment the basement membrane in the range of 0.70 ~ 0.85. The method can effectively segment the basement membrane whose thickness suddenly becomes narrower and trending to a sudden change, and obtain a relatively complete basement membrane.
【作者单位】：南方医科大学生物医学工程学院;南方医科大学基础医学院电子显微镜室;
【基金】：国家自然科学基金青年科学基金(61102114);国家自然科学基金(61372063)资助
【分类号】：R692.6

【相似文献】