A Comparative Evaluation of Deep Learning Methods in Automated Classification of White Blood Cell Images

Abstract

The identification and characterization of a patient’s blood sample are required for the diagnosis of blood-related disorders. As a result, the medical implications of automated methods for identifying and categorizing various kinds of blood cells are considerable. However, deep convolutional neural networks (CNN) and standard machine learning algorithms have performed well in the categorization of blood cell pictures. Red, White, and Platelets are all types of blood cells. Leucocyte, commonly known as the immune cell, is a type of blood cell that plays a vital part in human immune function. Depending on shape info and granulated data in leukocytes, white blood cells are usually split by hematologists into two different categories: non-granular cells (lymphocytes and monocytes) and granular cells (eosinophils, basophils, and neutrophils). The CNN portion receives the pre-trained weight parameters from the image dataset using the transfer learning approach. Also, We have used two different scenarios, the first scenario of using CNN directly gave us pictures. used SVM in the second scenario. Then we compare the best category results. The classification results demonstrated that the accuracy of CNN is 98.4 %, whereas the accuracy of Support Vector Machine (SVM) is 90.6 %. Other classifiers can be added to the suggested system to improve its performance.