Automated segmentation of biological cell boundaries and nuclei in highly Inhomogeneous and low contrast touching cell images

Abstract

Cells are the basic building blocks of human life. Cell image analysis has gained significant importance in the present era. Automatic image analysis is generally a difficult problem due to the large variability and complexity of the image data. Two key tasks in automated cell analysis are preprocessing and segmentation. Pre- processing refers to reducing the effect of image artifacts and segmentation is the process of locating the boundaries in cell images. Cell segmentation has gained significant importance in modern biological cell anal- ysis applications. The commonly used image segmentation algorithms are region based and depend on the homogeneity of the intensities of the pixels in the regions of interest. But owing to the highly inhomogeneous behavior of the cell nuclei and background, there is a feature overlapping between the two regions which leads to poor segmentation results. The present work has proposed a method to improve the homogeneity of the cell images. As touching cell is a major challenge in cell segmentation, a method has been pro- posed to segment the cell boundaries and cell nuclei in very low contrast touching cell images. Specific characteristics of low contrast cell images make the cell bound- ary detection more challenging. First, contrast of the cell images has been improved by a combination of multiscale top hat filter and h−maxima. Then, a curvelet based initialization of level set method has been proposed to initialize the modified Chan Vese level set method. As curvelets are capable of detecting image information along curves, extracting the curvelet coefficients at the desired scales and directions ini- tializes the level set model of the cell images more accurately. Further, a preventionterm has been added in the level set evolution equation to separate the touching cells. Finally, an attempt has been made to improve the speed of the Chan-Vese model for faster cell segmentation. The homogeneity improvement and cell segmentation results have been tested on different datasets. The results indicate an improved performance of the proposed algorithm.