The Segmentation Of Overlapping Cells In Cervical Smears Using Optical Absorption Model

Cervical cancer is one of the most common malignancies, which seriously endanger women's health. Early screening for cervical cancer is important to reduce the incidence and control of its mortality. Currently, liquid-based cytology test technology(LCT) is the most widely used screening technology for its high quality and quick production. Manual reading of smears is still used in most of hospitals, which is inefficient and easily influenced by the people's subjective factors. It is meaningful to use machine to read smear instead of manual film-reading by using the computer-aided diagnosis system, which can improve the efficiency and accuracy.The automatic segmentation of cell image is the basis of the computer-aided diagnosis system, whose accuracy has direct consequences on the quantitative analysis and classification of the subsequent steps. The segmentation of the cervical cell image consists of nuclei and cytoplasm. The single cell exist independently is relatively small and most cells have overlapping and adhesion phenomena in the images. Therefore, the segmentation of the overlapping cells is the key to the cytoplasm division, which is of great significance to improve the accuracy of the computer-aided diagnosis system.The main work of this paper is focused on the segmentation of the overlapping cervical cell image. Firstly, the preprocess divides the image into a series of homogeneous image block, which is convenient to the subsequence as well as reduce the calculate amount. Then, the translucency light transmission model proposed in the paper is applied to the segmentation of the overlapping, which can provide the theoretical foundation and important features. For the segmentation of the two overlapping cells, we design a target function combined with the overlapping color, colour uniformity, shape and the smoothness of the contour. And an optimized algorithm is proposed to complete the segmentation of the overlapping cells based on the target function. In order to solve the segmentation of two and more overlapping cells, we proposed a non-overlapping area construction algorithm and an optimized algorithm for the initial non-overlapping area by using the translucency light transmission model and the spatial relationship of cells. Moreover, we reconstruct the cell overlapping matrix based on the optimized non-overlapping area. Finally the segmentation result of the overlapping cells is generated by using the cell overlapping matrix and the optimized non-overlapping areas. To prove the efficiency of the algorithm, we evaluate the performance by using zsi, precision and recall. Experiment results indicate that the segmentation method has a good effect.