Using deep learning for blood cells detection

Abstract

The complete blood count is one of the most important and most performed exams in the medical field. Through it, it is possible to discover important alterations in the organism and it is used in the assessment of patients health. Although it is a common practice, the performance of exams is difficult in laboratories because of the high cost of purchase and maintenance machinery. As an alternative to that, this project develops a computational model of Deep Learning capable of detecting cells automatically through images of blood samples. Through the use of object detection libraries, it was possible to train a model aimed at this problem and capable of detecting cells in images with satisfactory precision. Considering the identification of cells in images of blood samples in the best results obtained, it was possible to count white cells with 100% accuracy, red cells with 89% and platelets with 96%, generating subsidies to elaborate the main parts of a blood count. The elements aimed at classifying types of white cells were not carried out due to the limitations of the dataset used. However, as it showed good results, the research can be expanded to future works that address this problem.