神经元干细胞序列图像的结合局部灰度阈值的水平集分割算法

Level-Set Segmentation Algorithm Combined with Local Gray Threshold on Image Sequences of Neuron Stem Cells

在光学显微镜成像的神经元干细胞序列图像中,针对目标与背景的弱对比度及细胞粘连、团簇等问题,提出一种新的分割算法.该算法基于无需初始化的水平集算法,通过引入曲率项来加速收敛;为降低算法的复杂度,提出衡量范数能量作为水平集进化的终止条件;最后结合局部灰度阈值法进一步分割粘连细胞.将该算法应用于2组细胞图像序列共120帧图像的分割中,不但解决了时间序列图像成像时焦距偏移带来的分割难题,而且能够准确地分离粘连、团簇细胞,并保留细胞的形态特征和位置信息.统计结果表明,分割成功帧所占整个序列的百分率较分水岭算法、传统水平集分割算法提高了30%～40%.

In optical microscopy imaged time lapse of neuron stem ceils image sequences, there exist low contrast ratio between objects and background, adherent and clustered cells＇ interference. A novel algorithm is presented aiming to solve these problems. It is based on a level-set without the need of re-initialization. After curvature term is added in order to accelerate convergence, iteration terminating condition is changed to measure norm energy in order to decrease complexity. Local gray threshold is combined with the result of curve evolution for clustered cells＇ separation at last. The presented algorithm is applied in two sequence images of 120 frames. The segmented results show that the algorithm can not only solve the problem of focus excursion but also separate adherent and clustered cells successfully as well as keep cells＇ shape and location. After compared with watershed and traditional level-set algorithms, this algorithm can improve the success rate to 30%～40%.