Fast automatic segmentation of cells and nucleuses in large-scale liquid-based monolayer smear images

In automated cervical cancer screening, the segmentation to find the contours of cells and nucleuses is one of the most important challenges. Existing methods are computationally expensive and may not be effective for both normal and cancerous cases. In this paper, we propose a novel approach for the rapid segmentation of cells and nucleuses in partial image of large-scale liquid-based monolayer smears. The image is first enhanced by contrast limited adaptive histogram equalization (CLAHE) and a bilateral filter. Otsu's double thresholding is used to initialize cells contours. Afterwards, we propose a robust low-intensity segmentation (RLIS) approach to initializes the contours of nucleuses. In addition, we propose two optimizers: adaptive non-iterative active contours (ANAC) and active contour with discretization (ACWD). One of the optimizers is selected to optimize the initial contours of the cells and nucleuses. Experimental results show that our method is able to achieve above 85% in accuracy, sensibility and specificity for both normal and cancerous cases with time complexity of O(n). Implemented in Python (ver. 3.5) on a 3.4GHz clock rate CPU, our method takes less than 1 second to process an image of 2048×2048.