Sistema computacional para la detección de glioblastomas en resonancia magnética usando aprendizaje no supervisado
Cargando...
Fecha
2020
Autores
Medelo Ballesteros, Hebert
Espinosa-Castro, Jhon-Franklin
Rodríguez, Johel E.
Neira Borja, James
Diaz, Viviana Elizabeth
Gómez Félix, Gabriela
Chango Cando, Freddy
Mallitasig Panchi, Betty
Añez, Roberto
A. Fossi, Cleiver
Título de la revista
ISSN de la revista
Título del volumen
Editor
Saber UCV, Universidad Central de Venezuela
Resumen
La imagenología por resonancia magnética convencional o avanzada es fundamental en la valoración de diversos tipos de tumores, incluyendo los glioblastomas. Ante la gran heterogeneidad e invasividad de estos, su manejo en la actualidad constituye un desafío complejo para los especialistas clínicos y neurocirujanos, así como para los desarrolladores de sistemas computacionales orientados al apoyo diagnóstico y de seguimiento. El presente sistema se utiliza para localizar y visualizar los glioblastomas en un conjunto de 11 bases de datos de uso público, incluye un procedimiento para la valoración de ciertas técnicas de filtrado en el realce de imágenes de resonancia magnética e integra la técnica de crecimiento de regiones de enlace sencillo como procedimiento de segmentación de los volúmenes filtrados. Los filtros Gaussiano, mediana y erosión se asumen como técnicas de realce del contraste asociado a las imágenes de resonancia magnética. El proceso de valoración comienza por el establecimiento de un conjunto de configuraciones de los filtros el cual depende de los valores y de las combinaciones de sus parámetros de entradas. El conjunto de configuraciones establecido es valorado a partir de la cuantificación de una medida de la calidad del realce, la cual se basa en una función de puntuación que agrupa un conjunto de métricas de mejora previamente reportadas. A partir del resultado de la valoración se concluye que el filtro con mejor desempeño corresponde con la mediana y se construye una interfaz gráfica de usuario para administrar la interacción entre el usuario y los algoritmos de procesamiento de imágenes.
Conventional or advanced magnetic resonance imaging is fundamental in the evaluation of several types of tumors, including glioblastomas, given the great heterogeneity and invasiveness of these tumors, their management at present constitutes a complex challenge for clinicians and neurosur- geons. This is also true for developers of computer systems oriented to diagnostic support and monitoring. The present system is used to localize and visualize glioblastomas in 11 datasets of public use. It includes a procedure for the evalu- ation of certain filtering techniques in the enhancement of magnetic resonance images and, furthermore, it integrates the technique of region growth and simple link as a segmen- tation procedure for filtered volumes. Gaussian, median, and erosion filters are assumed as contrast enhancement tech- niques associated with magnetic resonance imaging. The evaluation process begins with the establishment of a set of filter configurations which depends on the values and com- binations of their input parameters. The established set of configurations is valued from the quantification of a measure of the quality of the enhancement, which is based on a scor- ing function that groups together a set of previously reported improvement metrics. From the result of the evaluation, it is concluded that the filter with the best performance corre- sponds to the median. In addition, a graphic user interface is constructed to manage the interaction between the user and the image processing algorithms.
Conventional or advanced magnetic resonance imaging is fundamental in the evaluation of several types of tumors, including glioblastomas, given the great heterogeneity and invasiveness of these tumors, their management at present constitutes a complex challenge for clinicians and neurosur- geons. This is also true for developers of computer systems oriented to diagnostic support and monitoring. The present system is used to localize and visualize glioblastomas in 11 datasets of public use. It includes a procedure for the evalu- ation of certain filtering techniques in the enhancement of magnetic resonance images and, furthermore, it integrates the technique of region growth and simple link as a segmen- tation procedure for filtered volumes. Gaussian, median, and erosion filters are assumed as contrast enhancement tech- niques associated with magnetic resonance imaging. The evaluation process begins with the establishment of a set of filter configurations which depends on the values and com- binations of their input parameters. The established set of configurations is valued from the quantification of a measure of the quality of the enhancement, which is based on a scor- ing function that groups together a set of previously reported improvement metrics. From the result of the evaluation, it is concluded that the filter with the best performance corre- sponds to the median. In addition, a graphic user interface is constructed to manage the interaction between the user and the image processing algorithms.
Descripción
Palabras clave
Glioblastoma, Resonancia magnética, Realce de imágenes, Filtro de suavizado, Segmentación, Aprendizaje no supervisado, Visualización tridimensional, Interfaz gráfica de usuario, Magnetic resonance, Image enhancement, Smoothing filter, Segmentation, Unsupervised learning, Three-dimensional visualization, Graphical user interface