Desenvolvimento de algoritmos para avaliação da imagem do Simulador Phantom mama para o controle de qualidade em mamografia

Abstract

The most suitable test for the diagnosis of breast diseases is the mammography. In order to guarantee that the mammographic diagnosis is done properly and efficiently, the State Secretariat of Health of Minas Gerais created the State Program for Quality Control in Mammography. One purpose of the program is to monitor the image quality monthly, which consists of evaluating visually the images obtained using a breast phantom. With the visual analysis it is not possible to expand the program because of the need of trained personnel to carry out this evaluation. The goal of this study was to develop and validate algorithms for automatic evaluation of the masses, fibers and micro-calcifications present in Phantom mam, recognized by the Brazilian College of Radiology; and evaluate them in accordance with the visual assessment performed by humans. The algorithms were developed using films from database Monthly Monitoring Program acquired in conventional systems, digital CR and DR, that were later adjusted for images obtained in the mammography unit with CR system installed on LARAM/CDTN. These images were obtained using different parameters of Voltage, Load, combination Anode/Filter. The algorithms were developed in MATLAB® and evaluations of test objects were made following the parameters given by the manufacturer. For the detection of structures, selected regions of interest were selected in which preprocessing,segmentation techniques and morphological operators for post processing were applied. Data of the two evaluations (visual and automatic) were compared and it showed acceptable concordances. Through automatic evaluation, it was possible to reduce thesubjectivity caused by human evaluation. It was concluded that it will be possible to perform automatic evaluation with the developed algorithms, evaluate the image quality on mammography services and expand the Quality Control Program