Clasificación morfológica de eritrocitos en imágenes digitales de frotis de sangre periférica mediante deep learning

n hematology, the hemogram is one of the evaluative tests used with greater regularity in medical practice, since it allows to evaluate and quantify the different types of cells present in the blood. However, not all characteristics of blood cells can be detailed with this test, which is why a micro...

Full description

Saved in:

Bibliographic Details
Main Author:	Mena Quintero, María Camila
Other Authors:	Narváez Semanate, José Luis
Format:	Trabajo de grado (Pregrado y/o Especialización)
Language:	spa
Published:	Universidad Antonio Nariño 2022
Subjects:	Método de clasificación red neuronal Deep Learning clasificación morfológica Classification method erythrocytes morphological classification neural network
Online Access:	http://repositorio.uan.edu.co/handle/123456789/5972
Tags:	Add Tag No Tags, Be the first to tag this record!

Description
Summary:	n hematology, the hemogram is one of the evaluative tests used with greater regularity in medical practice, since it allows to evaluate and quantify the different types of cells present in the blood. However, not all characteristics of blood cells can be detailed with this test, which is why a microscopic inspection of the peripheral blood smear is required. The manual exploration of the blood smear, allows to extract, among others, qualitative information about the blood cells, by means of a visual inspection with the help of the microscope; The inspection is a detailed and orderly process, which is carried out with the aim of looking for morphological changes that make it possible to establish differences between normality and abnormality. Since it is carried out manually, the results of this type of classification, based on qualitative parameters; they depend on the skill and experience of the evaluator, which can lead to mistakes, time and money. Taking into account the aforementioned, an erythrocyte classification method was implemented in Matlab, based on morphological descriptors (diameter, perimeter, area, solidity, circularity and concavity), from which a neural network was trained, from which a percentage of accuracy of 83.3% is obtained.