Automatic White Blood Cell Classification Using the Combination of Convolution Neural Network and Support Vector Machine

Leukocyte tests can help diagnose several diseases related to blood disorders such as acute leukemia, chronic myeloid leukemia. Manual blood cell classification is commonly used in clinics and hospitals which, although is accurate, is not time-effective and affects clinical workflow. Automated cell classification systems might assist clinicians to increase the efficiency and accuracy of diagnosis. However, most of the published papers classify only non-pathological cells while applying a complex segmentation process. To overcome this problem, we proposed Convolutional Neural Network and Support Vector Machine (CNN-SVM) where Convolutional Neural Network model is used to extract features directly from the images. The obtained features will then be used by a following Support Vector Machine for classifying white blood cells into 5 different types include eosinophil, lymphocyte, monocyte, neutrophil, and pathological white blood cell (leukemia). Three different CNN models including Alexnet, Resnet-101, VGG-19 were tested on 15764 enhanced images to pick the best CNN model for extracting the features. Resnet-101 was chosen for its best average accuracy (97.8%) when combining with SVM to classify 5 white blood cell types. This average accuracy was also the second-best compared to other published methods from 2016 up to now.