An Automated Framework for White Blood Cells Classification and Segmentation using Optimised Deep Learning Techniques

Abstract

White Blood Cell (WBC) counts that are abnormal can be a sign of cancer, malignancy, newlinetuberculosis, severe anemia, and other dangerous illnesses. Examining the quantity and determining the morphology of the WBCs is necessary to make an early diagnosis and newlineto evaluate whether the white blood cells are abnormal or normal. Hematologists pursue newlinethis difficult, pricey, and time-consuming operation manually, thus computer-aided ap- newlineproaches have been created to solve this issue. As an outcome, in the present research, newlinea potent Deep Learning (DL) system was created to classify white blood cells, having newlineimmature white blood cells, from pictures of peripheral blood smears. For segmenta- newlinetion of leukocytes using W-Net, a CNN based technique for WBC categorization has newlinebeen created. A DL system based on GhostNet was then used to retrieve key feature newlinemaps. Then, a ResNeXt with a WHO-based algorithm was used to categorize them. To newlineaddress the issue of imbalanced data, DCGAN-based data enhancement was put into newlineplace. These results were used to validate the approach s performance. newlineIn the second proposed methodology, an automated approach for classifying and newlineidentifying WBCs from blood pictures is presented. The preprocessed blood cell pic- newlineture is then segmented utilizing SegNet, a powerful deep-learning framework. Then, newlineutilizing the EfficientNet design, the critical characteristics are created and retrieved.The XGBoost classifier is then used to divide the WBCs into four distinct groups: newlineeosinophils, neutrophils, lymphocytes, and monocytes. The proposed methods effec- newlinetiveness is assessed using the BCCD dataset, and the outcomes are contrasted with newlineprevious methods based on precision, accuracy, sensitivity, f-score, and specificity. newlineIn the third proposed methodology, a robust DL system for classifying WBCs based newlineon MobilenetV3-ShufflenetV2 is shown. Initially, a powerful Pyramid Scene Parsing newlineNetwork (PSPNet) is used to segment the WBC pictures. The local and global charac- newlineteristics are then extracted and chosen from the se