Multiscale information fusion to segment white blood cells for early stage detection of acute lymphoblastic leukemia
M. Thilagaraj, S. Vaira Prakash, G. Petchinathan, Kottaimalai Ramaraj, C. S. Sundar Ganesh, S. Krishnanarayanan · 2025
White blood cells (WBC) are most affected by the malignancy leukemia. Leukocytes, another name for WBCs, combat infestations. The squishy substance found within bones called the marrow of the bone is where RBC, WBC, and platelets are produced. The marrow in the bones produces ineffective WBCs when a patient has leukemia. The human system cannot be protected from pathogens by these aberrant cells. It overcrowds the bone marrow, get into the circulation of the blood, and may migrate to the cerebral cortex, liver, or lymph glands, among various regions of the human anatomy. If the immune system produces an excessive amount of lymphocytes, a kind of WBC, it can result in acute lymphoblastic leukemia (ALL). Among the deadliest forms of blood-based cancer, leukemia affects a large number of individuals annually. The detection of leukemia is significantly correlated with WBCs. Investigations have shown that leukemia alters the shape and quantity of WBCs. Effective WBC separation makes it possible to identify the quantity and morphological that subsequently aids in leukemia identification and treatment. Human WBC evaluation processes are laborious, arbitrary, and not as precise. To address challenges in White Blood Cell (WBC) categorization, researchers propose the utilization of a Multi-Scale Information Fusion Network (MIF-Net). This deep structure integrates both external and internal processes to fuse spatial information effectively. The difficulty in accurate segmentation of WBC images arises from the low contrast between the cytoplasm and the background, as well as the complex structure of nuclei with fuzzy borders. The early layers of the network focus on capturing precise boundary details as a spatial property. MIF-Net strategically di-vides and extends this boundary information across multiple scales to facilitate the fusion of external data. MIF-Net improves segmentation efficiency while protecting border details. Additionally, MIF-Net leverages internal information fusion at certain intervals to enhance features at various network phases.