A deep network designed for segmentation and classification of leukemia using fusion of the transfer learning models

Abstract White blood cells (WBCs) are a portion of the immune system which fights against germs. Leukemia is most common cancer may lead to death. It occurs due production large number immature WBCs in bone marrow that destroy healthy cells. To overcome severity this disease, it necessary diagnose shapes at an early stage ultimately reduces modality rate patients. Recently different types segmentation and classification methods presented based upon deep-learning (DL) models but still have some limitations. This research aims propose modified DL approach for accurate leukocytes their classification. The proposed technique includes two core steps: preprocessing-based segmentation. In preprocessing, synthetic images generated using generative adversarial network (GAN) normalized by color transformation. optimal deep features extracted from each smear image pretrained i.e., DarkNet-53 ShuffleNet. More informative selected principal component analysis (PCA) fused serially morphological operations on thresholding with semantic method utilized leukemia classified accuracy achieved ALL-IDB LISC dataset 100% 99.70% blast, no basophils, neutrophils, eosinophils, lymphocytes, monocytes, respectively. Whereas 99.10% 98.60% average global accuracy, outstanding outcomes as compared latest existing works.