Indian Journal of Science and Technology

Abstract

Objectives: To help doctors and hematologists in the Differential Blood Count process, in order to increase productivity and eliminate human errors. Methods: The automation of the Differential Blood Count process offers a low-cost solution, compared to high-tech medical equipment. Due to the multiple nature of these cells and the uncertainty in the hematological images, leukocyte segmentation is one of the most important stages in this process. Scrupulous segmentation obviously reduces the errors of the following stages. In this article, we present the K-means clustering algorithm in the Hue – Saturation - Intensity (HIS) color space to segment the cores. In addition, the performances of three classifiers, Support Vector Machine (SVM), Linear Discriminant Analysis (LDA) and Quadratic Discriminated Analysis (QDA) for the recognition of leukocyte types are compared. Findings: In the evaluation process, the technique proposed individually is applied to each of 147 blood smear images; 139 of them were segmented with precision, reaching an average precision of 94.6%. The test consists of classifying 52 leukocytes present in images obtained in the María Auxiliadora health center, during two sessions, which contains 14 lymphocytes, 12 monocytes, 8 eosinophils and 18 neutrophils previously classified by the bacteriologist. For lymphocytes, monocytes, eosinophils and neutrophils an accuracy of 98.1%, 90.4%, 92.3% and 88.5%, respectively, is achieved. Improvement: The application of the proposed method shows a 92.3% accuracy of the system to classify the cells.

Keywords: Discriminant Analysis (LDA), Hue – Saturation - Intensity (HIS), K-means, Leukocyte, Quadratic Discriminated Analysis (QDA), Support Vector Machine (SVM)