Clasificación de imágenes médicas para el diagnóstico de neumonía mediante técnicas de Machine Learning
Loading...
Official URL
Full text at PDC
Publication date
2023
Defense date
07/2023
Authors
Advisors (or tutors)
Editors
Journal Title
Journal ISSN
Volume Title
Publisher
Citation
Abstract
La neumonía es una de las principales causas de muerte tanto en niños como en personas mayores. En concreto, la neumonía infecta a los alvéolos llenándolos de pus y líquido, provocando que la respiración del paciente sea vea debilitada. Los pacientes con enfermedades respiratorias previas, sistema inmunitario débil, bebés hospitalizados o personas mayores con respiradores son más propensos a sufrir esta enfermedad, por lo que un diagnóstico temprano y eficiente resulta esencial para poder aplicar tratamientos adecuados.
En el presente trabajo, se abordará el problema de la detección de la neumonía mediante el uso de las imágenes de radiografías de tórax. Han sido propuestos el desarrollo y evaluación de modelos de aprendizaje profundo basados en redes neuronales convolucionales para diagnosticar las imágenes rayos X en dos categorías: neumonía y sano. Se llevó a cabo un preprocesamiento de las imágenes con el fin de mejorar la calidad de los datos para optimizar el rendimiento del modelo. Se experimentó con diferentes arquitecturas de red, tanto de modelos propios personalizados como modelos preentrenados usando técnicas de transfer learning. Se hará una comparativa entre todos los modelos implementados y se elegirá el más óptimo considerando diferentes factores, siendo el más importante el rendimiento de los modelos.
Los resultados obtenidos en este trabajo servirán como herramientas para acelerar los diagnósticos médicos, sirviendo como apoyo en el proceso de toma de decisiones y mejorando la eficiencia en la detección temprana de la neumonía. Esta herramienta siempre ha sido implementada con la intención de apoyar al personal médico para mejorar la precisión del diagnóstico, nunca con el objetivo de sustituirlos.
Pneumonia is a leading cause of death in both children and older individuals. Specifically, pneumonia infects the alveoli, filling them with pus and fluid, thereby weakening thepatient’s breathing. Patients with previous respiratory diseases, weakened immune systems, hospitalized infants, or older individuals on respirators are more prone to this disease, so making early and efficient diagnosis is essential for applying appropriate treatments. In the present study, the problem of pneumonia detection will be addressed through the use of chest X-ray images. The development and evaluation of deep learning models based on convolutional neural networks to diagnose X-ray images into two categories: pneumonia and healthy, have been proposed. Image preprocessing was carried out to improve data quality and optimize model performance. Different network architectures were experimented with, including custom models and pretrained models using transfer learning techniques. A comparison will be made between all implemented models, and the most optimal one will be chosen considering various factors, with the most important being the models’ performance. The result achieved in this study serves as an additional tool for medical diagnostic environments, offering support in decision-making and improving the efficiency of early pneumonia detection. This tool has always been implemented with the intent of supporting medical personnel to enhance the accuracy of medical diagnosis, never with the objective of replacing them.
Pneumonia is a leading cause of death in both children and older individuals. Specifically, pneumonia infects the alveoli, filling them with pus and fluid, thereby weakening thepatient’s breathing. Patients with previous respiratory diseases, weakened immune systems, hospitalized infants, or older individuals on respirators are more prone to this disease, so making early and efficient diagnosis is essential for applying appropriate treatments. In the present study, the problem of pneumonia detection will be addressed through the use of chest X-ray images. The development and evaluation of deep learning models based on convolutional neural networks to diagnose X-ray images into two categories: pneumonia and healthy, have been proposed. Image preprocessing was carried out to improve data quality and optimize model performance. Different network architectures were experimented with, including custom models and pretrained models using transfer learning techniques. A comparison will be made between all implemented models, and the most optimal one will be chosen considering various factors, with the most important being the models’ performance. The result achieved in this study serves as an additional tool for medical diagnostic environments, offering support in decision-making and improving the efficiency of early pneumonia detection. This tool has always been implemented with the intent of supporting medical personnel to enhance the accuracy of medical diagnosis, never with the objective of replacing them.