Estrategia basada en realce por similaridad local para la segmentación computacional de la vena cava superior en imágenes de tomografía computarizada cardiaca
No hay miniatura disponible
Fecha
2017
Autores
Vera, Miguel
Huérfano, Yoleidy
Contreras-Velásquez, Julio
Bermúdez, Valmore
Del Mar, Atilio
Cuberos, María
Título de la revista
ISSN de la revista
Título del volumen
Editor
Cooperativa servicios y suministros 212518 RS
Resumen
El artículo propone una estrategia para segmentar
la vena cava superior (VCS) en 20imágenes
tridimensionales (3-D) de tomografía
computarizada multicapa, correspondientes al ciclo cardiaco
completo de un paciente. Esta estrategia consta de
las etapas de pre-procesamiento, segmentación y entonación
de parámetros. La etapa de pre-procesamiento se
divide en dos fases. En la primera, denominada fase de
filtrado, se emplea una técnica denominada realce por
similaridad local (LSE) con el propósito de disminuir el impacto
de los artefactos y atenuar el ruido en la calidad
de las imágenes. Esta técnica, combina un filtro promediador,
un filtro detector de bordes (denominado black
top hat) y un filtro gaussiano (GF). En la segunda, identificada
como fase de definición de una región de interés
(ROI), se consideran las imágenes filtradas, máquinas de
soporte vectorial de mínimos cuadrados e información a
priori para aislar las estructuras anatómicas que circundan
la VCS. Por otra parte, durante la etapa de segmentación
3-D se implementa un algoritmo de agrupamiento, denominado
crecimiento de regiones (RG), el cual es aplicado
a las imágenes pre-procesadas. Durante la entonación de
parámetros, de la estrategia propuesta, el coeficiente de
Dice (Dc) es utilizado para comparar las segmentaciones,
de la vena cava superior, obtenidas automáticamente, con
la segmentación de la VCS generada, manualmente, por
un cardiólogo. La combinación de parámetros que generó
el Dc más elevado considerando el instante de diástole se
aplica luego a las 19 imágenes tridimensionales restantes,
obteniéndose un Dc promedio superior a 0.88 lo cual indica
una buena correlación entre las segmentaciones generadas
por el experto cardiólogo y las producidas por la
estrategia desarrollada.
The article proposes a strategy to segment the superior vena cava (VCS) into 20-dimen- sion (3-D) images of multi-layer computed tomography, corresponding to the complete cardiac cycle of a patient. This strategy consists of the stages of pre-processing, segmentation and intonation of parameters. The pre-processing stage is divided into two phases. In the first phase, called the filtering phase, a technique called local similarity enhancement (LSE) is used in order to reduce the impact of the artifacts and to attenuate noise in the quality of the images. This technique combines an averaging filter, an edge detector filter (called black top hat) and a Gaussian filter (GF). In the second, identified as a region of interest (ROI) definition phase, we consider filtered images, least squares vector support machines and a priori information to isolate the anatomical structures that surround the VCS. On the other hand, during the 3-D segmentation stage a clustering algorithm, called region growth (RG), is implemented, which is applied to the preprocessed images. During the intonation of parameters, of the proposed strategy, the Dice coefficient (Dc) is used to compare the segmentations of the superior vena cava, obtained automatically, with the segmentation of the VCS generated manually by a cardiologist. The combination of parameters that generated the highest Dc considering the instant of diastole is then applied to the remaining 19 three-dimensional images, obtaining an average Dc higher than 0.88 which indicates a good correlation between the segmentations generated by the expert cardiologist and those produced by the strategy developed.
The article proposes a strategy to segment the superior vena cava (VCS) into 20-dimen- sion (3-D) images of multi-layer computed tomography, corresponding to the complete cardiac cycle of a patient. This strategy consists of the stages of pre-processing, segmentation and intonation of parameters. The pre-processing stage is divided into two phases. In the first phase, called the filtering phase, a technique called local similarity enhancement (LSE) is used in order to reduce the impact of the artifacts and to attenuate noise in the quality of the images. This technique combines an averaging filter, an edge detector filter (called black top hat) and a Gaussian filter (GF). In the second, identified as a region of interest (ROI) definition phase, we consider filtered images, least squares vector support machines and a priori information to isolate the anatomical structures that surround the VCS. On the other hand, during the 3-D segmentation stage a clustering algorithm, called region growth (RG), is implemented, which is applied to the preprocessed images. During the intonation of parameters, of the proposed strategy, the Dice coefficient (Dc) is used to compare the segmentations of the superior vena cava, obtained automatically, with the segmentation of the VCS generated manually by a cardiologist. The combination of parameters that generated the highest Dc considering the instant of diastole is then applied to the remaining 19 three-dimensional images, obtaining an average Dc higher than 0.88 which indicates a good correlation between the segmentations generated by the expert cardiologist and those produced by the strategy developed.
Descripción
Palabras clave
Tomografía, Vena cava superior, Realce por similaridad local, Segmentación, Tomography, Superior vena cava, Local similarity enhancement, Segmentation