Hernández Martínez, Henry AlbertoMazorca Espitia, Juan DavidAlonso Castro, Cesar Camilo2025-03-072025-03-072024-10-08http://hdl.handle.net/11349/93409En el contexto actual de avances tecnológicos, el diagnóstico temprano y preciso de enfermedades complejas como el glioblastoma multiforme (GBM), un tipo agresivo de tumor cerebral es fundamental para mejorar las oportunidades de tratamiento y supervivencia de los pacientes. La proliferación de imágenes médicas digitales, especialmente las imágenes de resonancia magnética (MRI) en formato DICOM, ofrece una rica fuente de datos para la identificación y análisis de dichas patologías. El problema abordado en este trabajo es la necesidad de métodos automatizados y precisos para la detección de glioblastomas en imágenes de resonancia magnética. Tradicionalmente, el análisis de estas imágenes ha dependido de la interpretación manual de los radiólogos, lo que puede ser un proceso subjetivo, lento y propenso a errores. El objetivo principal del proyecto es desarrollar un sistema de apoyo que utilice redes neuronales profundas para analizar y detectar automáticamente las características distintivas del glioblastoma en imágenes DICOM. La solución propuesta consiste en una aplicación con una interfaz gráfica desarrollada en Tkinter, la cual permite a los usuarios cargar y visualizar imágenes DICOM, realizar el preprocesamiento necesario, y aplicar un modelo de red neuronal convolucional entrenado para identificar y resaltar áreas sospechosas de tumor. La implementación de esta herramienta no solo busca aumentar la precisión en la detección de glioblastoma, sino también acelerar el proceso de diagnóstico y reducir la carga de trabajo de los profesionales médicos. Los resultados obtenidos hasta ahora han demostrado que el sistema es capaz de identificar características del glioblastoma con un grado de exactitud del 91,84 % y una precisión del 87,5 % en referencia a pacientes con tumor de glioblastoma en resonancias magnéticas T2 tomadas en plano axial. Durante las pruebas, el modelo de red neuronal generó predicciones acertadas en comparación con diagnósticos realizados previamente por expertos. Estos hallazgos sugieren que la aplicación tiene el potencial de ser una herramienta de apoyo valiosa en entornos clínicos.In the current context of technological advances, early and accurate diagnosis of diseases complex such as glioblastoma multiforme (GBM), an aggressive type of brain tumor is essential to improve treatment opportunities and patient survival. The proliferation of digital medical imaging, especially magnetic resonance imaging (MRI) in DICOM format, offers a rich source of data for the identification and analysis of these pathologies.The problem addressed in this work is the need for automated and precise methods for the Detection of glioblastomas on magnetic resonance images. Traditionally, the analysis of These images have depended on manual interpretation by radiologists, which can be a subjective, slow and error-prone process. The main objective of the project is to develop a support system that uses deep neural networks to automatically analyze and detect the distinctive features of glioblastoma on DICOM images.The proposed solution consists of an application with a graphical interface developed in Tkinter, which allows users to upload and view DICOM images, perform preprocessing necessary, and apply a trained convolutional neural network model to identify and highlight suspicious areas of tumor. The implementation of this tool not only seeks to increase the accuracy in detecting glioblastoma, but also speed up the diagnostic process and reduce the workload of medical professionals. The results obtained so far have shown that the system is capable of identifying characteristics of glioblastoma with a degree of accuracy of 91.84% and a precision of 87.5% in reference to patients with glioblastoma tumor on T2 MRIs taken in axial plane. During testing, the neural network model generated accurate predictions in comparison with diagnoses previously made by experts. These findings suggest that The app has the potential to be a valuable support tool in clinical settings. In the current context of technological advances, early and accurate diagnosis of diseases complex such as glioblastoma multiforme (GBM), an aggressive type of brain tumor is essential to improve treatment opportunities and patient survival. The proliferation of digital medical imaging, especially magnetic resonance imaging (MRI) in DICOM format, offers a rich source of data for the identification and analysis of these pathologies.The problem addressed in this work is the need for automated and precise methods for the Detection of glioblastomas on magnetic resonance images. Traditionally, the analysis of These images have depended on manual interpretation by radiologists, which can be a subjective, slow and error-prone process. The main objective of the project is to develop a support system that uses deep neural networks to automatically analyze and detect the distinctive features of glioblastoma on DICOM images. The proposed solution consists of an application with a graphical interface developed in Tkinter, which allows users to upload and view DICOM images, perform preprocessing necessary, and apply a trained convolutional neural network model to identify and highlight suspicious areas of tumor. The implementation of this tool not only seeks to increase the accuracy in detecting glioblastoma, but also speed up the diagnostic process and reduce the workload of medical professionals.The results obtained so far have shown that the system is capable of identifying characteristics of glioblastoma with a degree of accuracy of 91.84% and a precision of 87.5% in reference to patients with glioblastoma tumor on T2 MRIs taken in axial plane. During testing, the neural network model generated accurate predictions in comparison with diagnoses previously made by experts. These findings suggest that The app has the potential to be a valuable support tool in clinical settings.Redes neuronales convolucionalesDICOMProcesamiento de señales digitalesExtracción de característicasDiagnóstico médico por imágenesbachelorThesisIngeniería en Control -- Tesis y disertaciones académicasGlioblastoma multiformeResonancia magnética en imágenesDiágnosticoRed neuronalConvolutional neural networksDICOMDigital signal processingFeature extractionMedical diagnostic imagingAbierto (Texto Completo)