| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 1.65 MB | Adobe PDF |
Authors
Advisor(s)
Abstract(s)
O cancro renal surge como um dos cancros mais comuns a nível mundial, apresentando elevada taxa de mortalidade. Esta tem vindo a diminuir graças aos avanços da tecnologia de imagem e diagnóstico localizado, que permitem detetar precocemente algumas lesões renais, que anteriormente nunca o seriam, bem como auxiliar os médicos na tomada de decisões acerca do tratamento ideal e na avaliação sistemática das terapias. Uma ferramenta poderosa para tal são as Imagens de Ressonância Magnética uma vez que proporcionam alto contraste. Além disto, a heterogeneidade, bastante característica dos cancros sólidos, condiciona o uso da biópsia, bem como a sua avaliação visual. Por outro lado, permite a conceção de imagens quantitativas, que posteriormente, serão processadas de modo a extraírem-se características, como textura, arestas e forma, identificadoras das lesões.
Esta dissertação teve como objetivo encontrar biomarcadores característicos de doenças renais com recurso a técnicas de processamento de imagem. Numa primeira fase realizou-se uma segmentação inicial dos rins, com recurso à equalização do histograma e difusão anisotrópica. A segunda fase teve como foco principal a extração de características. Para tal, foram realizadas filtragens passa-banda, filtragens gaussianas, cálculos das curvaturas médias, bem como de momentos estatísticos. De seguida, já com as características extraídas para cada pixel da imagem foram construídos os vetores descritores. Por último, foi realizada a segmentação da área renal com problemas patológicos baseada na classificação de cada pixel da região de interesse, usando Support Vector Machine.
Os resultados finais revelam uma segmentação bastante efetiva, em termos gerais, resultando em bons resultados de precisão (82.5%), de acurácia (98.6%), de sensibilidade (81.2%) e do coeficiente de dice (0.8), confirmando a metodologia proposta como sendo bastante promissora para definir biomarcadores característicos de doenças renais.
Renal cancer appears as one of the most common cancers worldwide, with a high mortality rate. This has been decreasing do to advances in imaging technology and localized diagnosis, which allow early detection of some renal lesions, impossible to achieve in recent years, improving decision making by physicians on optimal treatment and systematic evaluation therapies. A powerful tool is the Magnetic Resonance Imaging as it provides high contrast. Furthermore, the heterogeneity, which is quite characteristic of solid cancers, limits the use of biopsy, as well as its visual evaluation. However, it allows the design of quantitative images that can be processed in order to extract lesion identifiers characteristics, such as texture, margins and shape. This dissertation aimed to find biomarkers characteristic of renal diseases using image processing techniques. In the first phase an initial segmentation of the kidneys was performed using histogram equalization and anisotropic diffusion. The second phase had as main focus the extraction of characteristics. For this, band-pass filtering, gaussian filtering, calculations of mean curvatures, as well as statistical moments were performed. Then, using each pixel of the image extracted characteristics, descriptor vectors were constructed. Finally, for pathological segmentation, a pixel classification was performed using Support Vector Machine. In general terms, the segmentation procedure results in good precision (82.5%), accuracy (98.6%), sensitivity (81.2%) and dice coefficient (0.8). In summary, the final results confirm the proposed methodology as promising to define biomarkers characteristic of renal diseases.
Renal cancer appears as one of the most common cancers worldwide, with a high mortality rate. This has been decreasing do to advances in imaging technology and localized diagnosis, which allow early detection of some renal lesions, impossible to achieve in recent years, improving decision making by physicians on optimal treatment and systematic evaluation therapies. A powerful tool is the Magnetic Resonance Imaging as it provides high contrast. Furthermore, the heterogeneity, which is quite characteristic of solid cancers, limits the use of biopsy, as well as its visual evaluation. However, it allows the design of quantitative images that can be processed in order to extract lesion identifiers characteristics, such as texture, margins and shape. This dissertation aimed to find biomarkers characteristic of renal diseases using image processing techniques. In the first phase an initial segmentation of the kidneys was performed using histogram equalization and anisotropic diffusion. The second phase had as main focus the extraction of characteristics. For this, band-pass filtering, gaussian filtering, calculations of mean curvatures, as well as statistical moments were performed. Then, using each pixel of the image extracted characteristics, descriptor vectors were constructed. Finally, for pathological segmentation, a pixel classification was performed using Support Vector Machine. In general terms, the segmentation procedure results in good precision (82.5%), accuracy (98.6%), sensitivity (81.2%) and dice coefficient (0.8). In summary, the final results confirm the proposed methodology as promising to define biomarkers characteristic of renal diseases.
Description
Keywords
Biomarcadores Classificação Descritores Doenças Renais Imagem de Ressonância Magnética Segmentação de Imagem