Leukocyte segmentation and SVM classification in blood smear images

Automated leukocyte detection, segmentation, and classification is an important task in clinical diagnosis. In this paper we present an approach to leukocyte cytoplasm and nucleus segmentation that is robust with respect to image quality and cell appearance. Cell properties are described by a set of statistical color and shape features. Pairwise coupling of SVM classification results is used to determine cell type probabilities. Evaluation of the method on a set of 1166 images containing 13 different cell types has resulted in 95% correctly segmented cells and a classification accuracy of 88% (at 20% reject rate).