Realce de Candidatos a Nódulo Pulmonar en Radiografías de Tórax por Medio de Filtros de Convergencia
Fecha
Autor corporativo
Título de la revista
ISSN de la revista
Título del volumen
Editor
Universidad Distrital Francisco José de Caldas
Compartir
Director
Altmetric
Resumen
Descripción
Este artículo presenta el desarrollo de una herramienta de software para el realce de áreas sospechosas de nódulos pulmonares en radiografías de tórax. Los nódulos pulmonares son lesiones que se encuentran dentro de la regíon pulmonar y son los primeros indicadores de la presencia de un tumor cancerígeno en este órgano. En primera instancia, se implementa un método de segmentación automático. Luego, para el realce de los candidatos, se usan filtros de convergencia que evalúan el grado en el que converge el área circundante al píxel de interés, asumiendo que los posibles candidatos a nódulos presentan forma circular. En este trabajo se implementan cuatro filtros de convergencia: Moneda, Iris, Anillo Adaptativo y Banda Deslizante. Los filtros de moneda y de anillo adaptativo obtuvieron mejores resultados en cuanto a detección y número de candidatos por imagen. Al aplicar la técnica de resta contralateral junto con los filtros basados en características geométricas tales como área, redondez y excentricidad, se logró disminuir el número de candidatos.
This paper describes the development of a method that highlights suspicious areas oflung nodules on chest radiographs. Lung nodules are lesions found in the lung region,which are the first indicators of the presence of cancerous tumors. First, an automaticsegmentation method is implemented. Next, to highlight the candidates, convergence filterswere used in order to evaluate the degree in which the surrounding area converged tothe pixel of interest, assuming a circular shape for potential nodule candidates. Four convergence filters were implemented: Coin filter, Iris filter, Adaptive Ring filter and SlidingBand filter. Coin and adaptive ring filters obtained better results in terms of detection andnumber of candidates per image, images from the database of the JSRT (Japanese Societyof Radiological Technology). Applying the contralateral subtraction technique togetherwith filters based on geometric features such as areas, roundness and eccentricity, it waspossible to reduce the number of candidates.
This paper describes the development of a method that highlights suspicious areas oflung nodules on chest radiographs. Lung nodules are lesions found in the lung region,which are the first indicators of the presence of cancerous tumors. First, an automaticsegmentation method is implemented. Next, to highlight the candidates, convergence filterswere used in order to evaluate the degree in which the surrounding area converged tothe pixel of interest, assuming a circular shape for potential nodule candidates. Four convergence filters were implemented: Coin filter, Iris filter, Adaptive Ring filter and SlidingBand filter. Coin and adaptive ring filters obtained better results in terms of detection andnumber of candidates per image, images from the database of the JSRT (Japanese Societyof Radiological Technology). Applying the contralateral subtraction technique togetherwith filters based on geometric features such as areas, roundness and eccentricity, it waspossible to reduce the number of candidates.
Palabras clave
Chest radiographs, lung nodules, segmentation, convergence index filters, Radiografía de tórax, nódulos pulmonares, segmentación, filtros de convergencia.