A New Method for Splitting Clumped Cells in Red Blood Images

Abstract

Automated cell counting is a required task which helps examiners in evaluating blood smears. A problem is that clumped cells usually appear in images with various degree of overlapping. This study presents a new method for effectively splitting clumped cells using value in distance transform of image to quickly detect central point. Additionally, a boundary-covering degree of each point is applied to select the best fit points. Another way to cell size estimation based on single cell extraction is also employed. With results from average cell size, central points with their boundary-covering degree, over-lapping cells in the image can be split correctly and rapidly. The robustness and effectiveness of our method have been assessed through the comparison with more than 400 images labeled manually by experts and exhibiting various clumped cell. As the result, the F-measure generally reaches 93.5% and more than 82% clumped cells can be tolerated in the condition of non-distorted shape and well-focused images.