No abstract

O controle de câncer de mama representa um dos grandes desafios que os serviços de saúde pública enfrentam atualmente. Para realizar o controle de qualidade de sistemas mamográficos o Ministério da Saúde exige o uso de simuladores radiográficos (phantoms) de mama. Com o objetivo de reduzir a subjetividade na avaliaçáo das imagens de phantom pela inspeçáo visual humana, está sendo desenvolvido um sistema computadorizado que utiliza um método de correçáo em imagens digitalizadas, associado à classificaçáo de suas estruturas de interesse pelo critério de visibilidade. Ao comparar os resultados da classificaçáo através do algoritmo J48 da ferramenta WEKA com e sem a correçáo das imagens, essa técnica apresentou uma melhora significativa na eficácia para determinadas estruturas do phantom.

This work corresponds to an extensive investigation on the effect of the image digitization process on an image processing scheme in mammography. As all the processing steps are based on a digitized image, different equipment, technologies, acquisition softwares and characteristics regarding such a procedure contribute to yield different features during the digitization of a same mammographic film. As a consequence, all the steps since the pre-processing up to the classification, mainly the images segmentation, could give different results depending on the digital image fidelity relatively to the original mammogram. Therefore, this research concentrates on a comparative evaluation of mammographic images acquired from several digitization systems, concerning the effect of this procedure on the sensitivity of a CAD scheme sections previously developed in our research group. A computer model is proposed in order to compensate some flaws introduced in the digitization procedure. The purpose is assuring more uniformity to the digital mammography images, no matter the digitizer used. Tests with digital images from several different systemssince common equipment, with transparency adaptors, up to the most expensive and sophisticated , with laser scanningby using the digitization driver proposed in this work have shown an increase in microcalcifications detection sensitivity for all cases relatively to the use of original digitized images without making use of the proposed model. Furthermore, the reduction in false positive rates was significant (from 70% up to 90%) in the same conditions. Therefore, the current application makes possible to any radiological institution the use of processing schemes to aid the detection and/or diagnosis of suspect structures in mammography, even using less sophisticatedand thus, with low costdigitizers and keeping the performance quality.