Improved leukocyte classification in bone marrow cytology using convolutional neural network with contrast enhancement

Leukocytes or white blood cells (WBCs) are the main components of the immune system that protect the human body from various infections caused by viruses, bacteria, fungi, and other microorganisms. There are five major types of leukocytes: basophils, lymphocytes, eosinophils, monocytes, and neutrophils. The precise identification and enumeration of each variety of WBCs are essential for the diagnosis and management of various conditions, including infectious diseases, immune disorders, immunological deficiencies, leukemia, and so forth. The conventional method of examining bone marrow cells by hematologists and pathologists using microscopy is tedious, time-consuming, and prone to variability among observers. Hence, there is a demand for a rapid and precise WBCs classification model. The proposed framework is highly accurate for the classification of leukocytes. A large dataset of leukocyte images was used in this study for training and testing. We used transfer learning to speed up the training process empowered with Contrast Limited Adaptive Histogram Equalization (CLAHE) technique to improve image quality and classification accuracy. The initial accuracy of the model was 81%. After the application of the CLAHE technique, the proposed approach significantly improved overall accuracy from 81 to 96.5% (15.5% improvement), outcompeting the state-of-the-art methods for leukocyte classification. Image contrast enhancement techniques, particularly CLAHE, improve the convolution neural network (CNN) model's performance. The proposed model can significantly assist hematologists and pathologists in accurately identifying leukocytes, thereby aiding in the detection of blood disorders and enabling more effective treatment strategies.