Convex Hull Based Detection of Overlapping Red Blood Cells in Peripheral Blood Smear Images

The Segmentation of Red Blood Cells (RBCs) in blood smear images to obtain their count is often the first step in the diagnosis of various pathological conditions. Although several procedures have been devised for this task, a majority of them suffer from performance degradation due to the overlapping of cells. Various researches have been carried out to split these overlapping cells. The proposed paper aims at suggesting two algorithms to find the concavity points in the overlapping RBCs’ contours. In the first approach, the dip points are obtained by analyzing the concave regions, obtained by finding out the Euclidean distance of all points in the overlapping cell to their convex hull. In the second approach, dip point identification is based only on the convex hull of the overlapping cell. The contours of the concave regions are analyzed from the perspective of the centroid. These two strategies were compared with the approach used in an earlier work, which also addressed the splitting of overlapping RBCs, by identifying the dip points using curve fitting and smoothing of the contours. The two approaches proposed in this paper are quite efficient in terms of accuracy and the time taken to achieve results. The specificity of the first approach was 90% and that of the second approach was 94%, meaning that the two new methods are far more advanced than the earlier work for which the specificity was only 75%.