A computer assisted method for leukocyte nucleus segmentation and recognition in blood smear images

The number or ratio of leukocytes in the blood is often an indicator of diseases. The leukocyte differential count is the process of diagnosing diseases by counting the number and ratio of leukocytes in the blood. However, leukocytes are usually manually classified in laboratories by using microscopes. It is a painstaking and subjective task for biologists. In order to improve the recognition accuracy and decrease the time consuming, an automatic leukocyte detective system is essential for assisting biologists in diagnosing diseases. This paper provides a method to detect and recognize leukocyte automatically. In general, leukocytes are categorized into five groups, including lymphocytes, monocytes, eosinophils, basophils and neutrophils. Each category plays a different role in the human immune system. The nucleus contains the main composition of a leukocyte and it can be used as an important feature to classify a disease. In this paper, the nuclei are used to identify five types of leukocyte. The leukocyte cell nucleus enhancer is proposed to segment the region we are interested in by enhancing the region of the leukocyte nucleus and suppressing the other region of the blood smear images. The segmental threshold of leukocyte nuclei is based on the Otsu's method. In addition, the erythrocyte size is estimated not only to assist the filtering out of the non-leukocyte objects but also to classify the leukocytes. Then the co-occurrence matrix is used as a textural measure of segmented images. In addition to the texture measure, we take into account of shape measures. In the recognition steps, we reduce features by principle component analysis (PCA) to obtain suitable feature to distinguish the five types of leukocytes. The genetic algorithm based k-means clustering approach is used to classify the five kinds of leukocyte in the reduced dimensions. The experimental results show that even though only leukocyte nucleus features are used for classification in our method, we achieve a high and promised accurate recognition rate.

► We propose a method to segment and recognize leukocyte nuclei automatically.
► A color-based enhancer is proposed to assist for further segmentation.
► The texture and shape features are used to classify the type of leukocytes.
► A genetic based k-mean clustering method is proposed to classify the type of leukocytes.