Research On Pathological Image Segmentation

Cervical cancer is a gynecological malignant tumor with high incidence and cervical smear screening is one of the effective methods to prevent cervical cancer.The segmentation of cer-vical cells is the core part of the automatic cervical smear analysis system while existing algo-rithms are slow and ineffective to handle overlapping cells.This thesis presents an effective automatic segmentation algorithm,especially for overlap-ping cells.An image is tranformed to polar coordinate based on the morphological character-istics of the cells,then a cost matrix is computed by a sampling step on the gradient image and optimized with the information of regions around the nuclei and the cytoplasm,finally a search step with a shape constraint is performed on the cost matrix and the global minimal cost path is found to obtain the cell boundary.The sampling method on polar coordinate can effectively reduce the solution space,making the algorithm relatively fast.The superpixel segmentation method can divide the image into many image blocks with uniform color and size and differ-ent objects are segmented into different superpixels.This thesis also designed a minimum cost path algorithm combined with the superpixel method,the ability of superpixel methods to ex-tract boundary information is fully used in the designed segmentation algorithm,the value of each pixel is replaced by the average value of the corresponding superpixel.The mean value images are used as the input images of the minimum cost path to obtain the segmentation result.Quantitative analyze is performed to measure the effect of the algorithms both in isolated cells and overlapping cells.Comparison of the designed method with RGVF segmentation algorithm and Level Set algorithm is also provided.The experiments are conducted on real cervical cell images and the results show that the designed algorithms can effectively solve the segmentation problem of overlapping cells with high speed,and can fulfill the actual needs of the automatic cervical smear analysis.