Low Cost Classification Method for Differentiated White Blood Cells using Digital Image Processing and Machine Learning Algorithms

Abstract

There are numerous commercially available technologies for counting blood components, however, these require expensive clinical equipment that is not affordable in most developing countries. In this manuscript, a low cost method for classification algorithm to calculate differentiated white blood cells through digital image processing, along with machine learning techniques is introduced. 98 images from ten (10) volunteers were used for this study. These samples were taken with the Venoject system; with a vein blood extraction, smear, and Wright staining to visualize differentiated white cells. The images were preprocessed to highlight their main characteristics (nucleus morphology, plasma membrane definition, cell color, etc.). These characteristics were labeled using the Bag of Visual Words (BoVW) method and classified using the Ensemble Subspace K Nearest Neighbor (ESkNN), Support Vector Machine (SVM), Linear Discriminant Analysis (LDA) and K Nearest Neighbor (KNN) models. Finally, data training and evaluation was performed. As a result, a database of peripheral blood smear pictures and an automatic counting system with 96.4% accuracy detection was attained.