Contour feature-point tagging as a mechanism for automated splitting of highly-occluded and dissimilar-sized cells in blood smear images

The presence of clumps in blood smear images may degrade the performance of automated cell counting techniques. This paper proposes a new system to split the occlusions in blood smear images. Prior art algorithms address problems like splitting of touching cells using concavity points and occlusion of only similar sized cells using distance transform. The system proposed in this paper addresses real life problems like heavily overlapped cells, dissimilar sized cell occlusions and uneven cell contours. The uniqueness of the paper lies in introducing geometrical measures in blood smear image analysis, tagging of feature points on the contour of the occluded segment, splitting of occlusions containing dissimilar sized cells and splitting of heavily overlapped cells. Using the feature points, approximate center of individual cells are calculated. To account for uneven cell contours, an iterative method is adopted to find the accurate center from the obtained approximate center. Using these accurate centers as reference, the occlusion is split into its constituent cells. The experimental results show that the system is highly robust and immune to cell deformations and heavily overlapped cells. A set of 12 images from different blood smear slides were taken for the study. The results show successful splitting for even 85 percent of overlaps.