White blood cell image classification using deep learning method

Abstract

White blood cells play an important role in the field of diagnosis of major blood related diseases. Our body contains mainly the three types of blood cells, which include red blood cells, white blood cells (WBCs), and platelets. WBCs are also called immune cells because they are the cells of the immune system. Identification of blood-related diseases are mainly based on the characteristics or the properties of the white blood cell's nucleus. So the reliable classification of white blood cells is important and increasingly demanded. Insufficient data is one of the limitation in the existing methods. Also unwanted background elements in the dataset reduced the classification performance. So that to classify the blood cell images, an enhanced method using convolutional neural network approach is proposed. The LISC database is used for the classification of the white blood cells. LISC is a public database which contain five classes of blood cell images. Some classes contain less number of images, so certain data augmentation techniques like flipping, rotation, shearing etc are applied. The augmented dataset contains irrelevant background along with white blood cells, so that the proposed method introduced a segmentation technique to remove the irrelevant portions. The images with unwanted background elements leads to low noise to signal ratio and which affect the classification performance. In this paper proposed a DenseNet architecture, which is trained to differentiate among five different classes of white blood cells. The performance of the system is evaluated based on the Accuracy, Precision and Recall achieved. Experimental results demonstrate that the proposed model can achieve better classification performance than traditional CNN. The best DNN model, DenseNet-201 yields an accuracy of 97.79%.