IX Simpósio de Engenharia Biomédica – SEB 2016 EVALUATION OF DIGITAL FILTERS FOR SIMILARITY ANALYSIS BETWEEN TOMOSYNTHESIS AND 2D MAMMOGRAPHIC IMAGES P. C. Carneiro*, G. D. Gallucci*, V. S. Rocha*, C. L. Debs**, N. Barros***, A. O. Andrade*, A. C. Patrocinio* *Faculty of Electrical Engineering, Federal University of Uberlândia, Uberlândia, Brazil. **Faculty of Medicine, Federal University of Uberlândia, Uberlândia, Brazil. ***Faculty of Medicine, University of São Paulo, São Paulo, Brazil. e-mail: pedrocarneiro@ufu.br Abstract: Breast cancer is the second leading cause of cancer death in women. Tomosynthesis is a new technique of additional exam that was created in order to improve the early detection of such disease. This technique shows a reconstructed image from different projections of the breast, reducing, this way, the effect of quantum noise due to the small thickness of the object of study. This paper aims at comparing the 3D reconstructed image with some filtered 2D polymethylmethacrylate (PMMA) images using structural similarity index (SSIM). We tested Wiener, Non-local Means, and Adaptive Median digital filters, which were applied at a region of interest of a 2D conventional mammography acquired in combo mode. From this work, we could quantify the level of similarity from different digital filters. Wiener and Adaptive Median filters increased the similarity between the 2D and the 3D tomosynthesis image, in terms of luminance, contrast and structure. However, the image with Non- local Means decreased significantly the similarity between the pair of compared images, proving the ineffective of this filter in reducing quantum noise. Keywords: 2D mammography, digital filter, similarity, SSIM, tomosynthesis. Introduction For the past few years, while aiming for an earlier detection of breast cancer, a new and promising technique of digital image acquisition has been created: tomosynthesis [1]. On this new technique, commonly referred to as 3D mammography, the X-ray tube is rotated in a single plan around the compressed breast, generating a series of projections, one for each angulation of the X-ray tube. Thus, many slices are produced, that is, images in thin cuts of the breast from the series of projections that are generated [2-3]. When using this system, tumors are identified more easily, specially in dense breasts, due to a higher level of detail provided by this type of image and a higher contrast variation compared to the conventional mammography 2D images [2, 4-6]. A way to enhance certain 2D mammograms characteristics is by using digital filters that remove undesirable signal, mainly the quantum noise, improving the image assessment [7-9]. The quality of the images can be assessed quantitatively with the help of several similarity indexes described in scientific literature [10-14]. The structural similarity (SSIM) algorithm is regarded as well consolidated. It is broadly accepted due to its great evaluation precision and simple mathematical formulation [14]. The aim of this paper is to quantitatively compare the 3D reconstructed image (used as reference) with filtered 2D polymethylmethacrylate (PMMA) images, evaluating the influence of the filters used, through the structural similarity index (SSIM). Materials e methods For the image acquisition, we used a Full Field Digital Mammography from Hologic Inc,, Selenia Dimensions model that has the tomosynthesis technology included for image acquisition. The images were acquired at Radiology Institute of the University of São Paulo using polymethylmethacrylate (PMMA) plates totaling 50mm of thickness. This equipment can produce the combined 3D and 2D images, known as combo mode, an imaging mode whereby both a 3D and 2D digital mammography image set are acquired in one breast compression. For the tests, we selected two images in the combo mode of the same exam (30 kVp): conventional 2D image and the 3D reconstruction generated from the 15 projections produced in the PMMA plates. The conventional 2D image is nothing more than a “projection” acquired with the X-ray tube positioned in a zero degree angle. After the image acquisition, the 2D image was the only one post-processed with digital filters, since it is known that this type of image is affected by quantum noise. Digital filtering consists in applying operators/masks which aims at softening, correcting or enhancing an image characteristic on a certain specific application, trying to reduce undesirable artifacts of the image.