Classification and segmentation of white blood cells using convolutional neural networks

In recent years, detection, classification and segmentation of white blood cells from blood smear images using deep learning methods have become quite common. Classification and segmentation of white blood cells plays an important role in the diagnosis of diseases such as leukemia, anemia and various infections. In recent years, due to the inadequacy of manual methods and traditional algorithms such as examining blood smear images under a microscope, deep learning approaches have been frequently used in solving such problems with the increase in processing power. In the presented thesis, two different artificial neural network models were created for the classification and segmentation of white blood cells. In classification, white blood cells are divided into 5 classes: basophil, neutrophil, eosinophil, monocytes and lymphocytes. In segmentation, it is divided into 4 classes (neutrophil, eosinophil, lymphocyte and monocytes) except basophils. The basophil cell type was not used for segmentation because its nucleus and cytoplasm were almost the same size. Semantic segmentation method was used for segmentation of white blood cells. Three pixel class labels were determined for segmentation: background, nucleus, and cytoplasm. Various performance evaluation metrics were used to evaluate the performance of the studies. Accuracy, precision, sensitivity and F1 Score values for classification; For segmentation, accuracy, Boundary F1 (BF) score and Intersection over Union (IoU) values were calculated. In the presented thesis study, it has been determined that the proposed method has better performance in classification when compared to other studies in the literature. In addition, with the successful results obtained with the segmentation algorithm, which is another proposed method, a diagnostic tool that can be used in the diagnosis of diseases in the clinical field creates a future perspective. In this presented study, it is important to study that it can play an important role in the early diagnosis of diseases, as it provides rapid and accurate classification and segmentation of white blood cells. In addition, by training the created system, it is aimed to create a system that can recognize these cells by detecting the parts with white blood cells from large microscope images.