Banca de DEFESA: RICARDO FELIPE FERREIRA

MANUAL DIFFERENTIAL COUNTING OF LEUKOCYTES USING IMAGE SEGMENTATION, ARTIFICIAL INTELLIGENCE AND MOBILE APPLICATION

Keywords: Leukocytes, count, white cells, differential count, artificial intelligence, convolutional networks, deep learning

Introduction: The analysis of blood cells began, through a simple microscope, in 1668. Since then, the sector has been improving until it reached automated counters, performing differential counting of leukocytes (white cells). The current automated blood count systems are large devices, and it is generally restricted to centralized laboratories and, in a laboratory routine, a differential leukocyte count must be performed manually, which generates a loss of efficiency and, consequently, of money for the labs. Objective: Within this context, the objective of this work is to analyze the technical and operational viability of a new product and methods that help health professionals who counts leukocytes. Methodology: A technical-marketing mapping and studies of algorithms that technically support the solution were carried out. Results: The trained models generated accuracy of 68.87% for lymphocytes, 100% for eosinophils, 99.36% neutrophils and 100% for monocytes for a test bed trained with a specific type of reagent and lens, for other types of conductor and lens, an accuracy of 70% was obtained. Conclusion: Concluding that there is technical and operational viability for a product that gives more dynamism and efficiency to blood test clinics, but there are challenges in the solidification of a process where the health professional includes a device or computer to perform a manual leukocyte count.