Segmentação da coluna vertebral humana por meio do processamento de imagens externas da região dorsal
Carregando...
Data
2018-03-02
Autores
Paulo, Jean Vitor de
Título da Revista
ISSN da Revista
Título de Volume
Editor
Universidade Estadual Paulista (Unesp)
Resumo
Neste trabalho avaliou-se a segmentação da coluna vertebral humana utilizando imagens externas da região dorsal. A avaliação foi realizada utilizando imagens de 70 pessoas (58 mulheres e 12 homens). Essas imagens foram agrupadas por meio da associação entre a quantidade de informação existente, dado pelo valor de entropia da imagem e uma avaliação qualitativa de visibilidade da musculatura paravertebral, realizada por três avaliadores. A segmentação foi feita utilizando um algoritmo, chamado DISLo (Dorsal Image Spine Locator), que processa imagens da região dorsal baseando-se na informação visível. Após o processamento, o algoritmo DISLo produz uma imagem binária contendo uma linha de pixels de intensidade 255 que representam a coluna vertebral identificada. Aplicando o algoritmo em todas as imagens, obteve-se uma segmentação de mais de 75% da coluna vertebral na maioria dos casos (40 imagens), e na minoria (4 imagens), menos de 25%. Posteriormente, para avaliar a qualidade da segmentação, utilizou-se o RMSD (Root Mean Square Deviation) calculado entre os pixels da segmentação automática do DISLo e outra realizada de modo manual, obtida da média de 9 avaliações realizadas por três avaliadores. Pôde-se verificar que as segmentações possuem uma exatidão maior em imagens com mais entropia, bem como possuem uma diferença no RMSD de +-2 pixels quando comparadas a imagens radiográficas. Portanto, a utilização de imagens externas da região dorsal para identificação da coluna vertebral é viável e apresenta uma maneira confiável para avaliação da mesma.
In this work, the segmentation of the human vertebral column was evaluated using external images of the dorsal region. The evaluation was performed using 70 individual images (58 women and 12 men). These images were grouped using the association between the amount of existing information, given by the entropy value of the image, and a qualitative assessment of the paravertebral musculature visibility, performed by three evaluators. The segmentation was performed using an algorithm called DISLo (Dorsal Image Spine Locator) which processes images from the dorsal region, based on the visible information. After the processing, the DISLo algorithm produces a binary image containing a line with pixels having a 255 intensity value that represents the identified backbone. Applying the algorithm to all images resulted in a segmentation of more than 75% of the spine in most cases (40 images), and in the minority (4 images), less than 25%. Subsequently, to evaluate the quality of the segmentation, the RMSD (Root Mean Square Deviation) was calculated between the pixels of DISLo's automatic segmentation and a manual one, obtained from the average of 9 evaluations performed by three evaluators. It could be verified that the segmentations have a greater accuracy in images with more entropy as well as having a difference in the RMSD of +-2 pixels when compared to radiographic images. Therefore, the use of external images of the dorsal region for identification of the spine is viable, and a reliable method for its evaluation.
In this work, the segmentation of the human vertebral column was evaluated using external images of the dorsal region. The evaluation was performed using 70 individual images (58 women and 12 men). These images were grouped using the association between the amount of existing information, given by the entropy value of the image, and a qualitative assessment of the paravertebral musculature visibility, performed by three evaluators. The segmentation was performed using an algorithm called DISLo (Dorsal Image Spine Locator) which processes images from the dorsal region, based on the visible information. After the processing, the DISLo algorithm produces a binary image containing a line with pixels having a 255 intensity value that represents the identified backbone. Applying the algorithm to all images resulted in a segmentation of more than 75% of the spine in most cases (40 images), and in the minority (4 images), less than 25%. Subsequently, to evaluate the quality of the segmentation, the RMSD (Root Mean Square Deviation) was calculated between the pixels of DISLo's automatic segmentation and a manual one, obtained from the average of 9 evaluations performed by three evaluators. It could be verified that the segmentations have a greater accuracy in images with more entropy as well as having a difference in the RMSD of +-2 pixels when compared to radiographic images. Therefore, the use of external images of the dorsal region for identification of the spine is viable, and a reliable method for its evaluation.
Descrição
Palavras-chave
Coluna vertebral, Processamento de imagens, Entropia, RMSD, Spine, Image processing, Entropy