Automatic classification of leukocytes using morphological features and Naïve Bayes classifier

In human body, different types of diseases can be found while examining blood samples. In this paper, the main aim is to detect leukocytes from human blood. These leukocytes protect the body from infectious diseases. If there is any type of disturbances in the blood count, then it may signify the presence of some cancer. Due to this reason, earlier many hematological experts examined blood samples by using medical instruments such as flow cytometer, for determining what type of disease is present in the human body. Since, the manual segmentation remains a very soporific, tiresome and error prone job, experts preferred to use automatic systems that results in accurate segmentation of leukocytes, which is needed for their classification. In this paper, the simple thresholding technique is used for segmentation of leukocytes by using Otsu thresholding. After segmentation, mathematical morphing is used to remove all components those do not look like leukocytes. Further, only the nucleus region was considered for feature extraction. Thereafter, Naïve Bayes classification technique is used for classification of leukocytes. The results obtained are better than other state of art algorithms, the classification accuracy on the training dataset of only 20 images and test image dataset of 68 images is about 80.88%, in an average time of 22 s per image.