电子测量与仪器学报

吴京城,洪欢欢,施露露,闻路红,杜亚南,史振志.反背景差分结合 Otsu 的细胞图像分割方法[J].电子测量与仪器学报,2021,35(4):82-89

反背景差分结合 Otsu 的细胞图像分割方法

Cell image segmentation method combined withanti-background subtraction and Otsu

DOI：

英文关键词:phase contrast microscope mesenchymal stem cells anti-background subtraction Otsu confluency

基金项目:国家重点研发项目(2018YFC1603504)、国家自然科学基金(81401452)、宁波市自然科学基金(2017A610164)资助

作者	单位
吴京城	1. 宁波大学高等技术研究院
洪欢欢	1. 宁波大学高等技术研究院
施露露	2. 宁波华仪宁创智能科技有限公司
闻路红	1. 宁波大学高等技术研究院
杜亚南	1. 宁波大学高等技术研究院
史振志	1. 宁波大学高等技术研究院

Author	Institution
Wu Jingcheng	1. The Research Institute of Advanced Technologies, Ningbo University
Hong Huanhuan	1. The Research Institute of Advanced Technologies, Ningbo University
Shi Lulu	2. China Innovation Instrument Co. , Ltd
Wen Luhong	1. The Research Institute of Advanced Technologies, Ningbo University
Du Yanan	1. The Research Institute of Advanced Technologies, Ningbo University
Shi Zhenzhi	1. The Research Institute of Advanced Technologies, Ningbo University

摘要点击次数: 392

全文下载次数: 5

中文摘要:

针对相差显微镜采集的间充质干细胞图像具有对比度低、背景不均匀、光晕伪影等问题,提出了反背景差分结合 Otsu 的细胞图像分割方法。该方法通过构建反背景差分增强图像中细胞主体与非细胞区域的差异,降低背景不均匀干扰因素的影响,结合 Otsu 阈值分割法粗略区分细胞和背景,并通过二值形态学运算、图像滤波和局部梯度迭代的算法组合进一步修正分割结果。通过对实际采集的细胞图像进行分割验证,像素精确度、交并比、Dice 相似性系数和汇合度误差四个评价指标分别达到了 0. 933 8、0. 729 6、0. 852 4 和 0. 07,表明该算法具有较高的分割性能,能客观、准确自动分析细胞汇合度,而且可以处理细胞不同培养时期的图像,具有较高的应用价值。

英文摘要:

Aiming at the problems of low contrast, uneven background and halo artifacts in the images of mesenchymal stem cells collected by the phase contrast microscope, this paper proposes a cell image segmentation method combined with anti-background subtraction and Otsu. The method constructs anti-background subtraction to enhance the difference between the cell body and the noncellular area and reduce the influence of uneven background, combines the Otsu threshold segmentation method to roughly distinguish the cells and the background, and further corrects the segmentation results by a combination of algorithms including binary morphology operations, image filtering, and local gradient iteration. The four evaluation indexes of pixel accuracy, intersection over union, dice similarity coefficient, and confluency error achieved values of 0. 933 8, 0. 729 6, 0. 852 4, and 0. 07, respectively, by segmentation validation on the actually acquired cell images. The results indicate that the algorithm has high segmentation performance, can objectively, accurately and automatically analyze the confluency of cells, and can process images of cells in different culture periods, which has high application value.

查看全文查看/发表评论下载PDF阅读器