Proposta de uma metodologia para suavização de ruído em imagens mamográficas de mamas densas

Abstract

Breast cancer is more common among women and is one of the biggest causes of death. Research shows that early diagnosis of the tumor may increase the chance of the disease occurring, and currently the most effective for screening and early detection is a mammogram. However, during the image acquisition process there is the insertion of noise into the signal, this is due to instrumentation, transmission errors and compression. Usually the noise is quantified by the percentage of signs that are corrupted and the most common types in mammography are quantum and gaussian. For this reason, use denoising processes to remove the noise present in the mammographic image becomes essential to obtain an image of better quality and thus facilitate the detection of some finding in the breast. This work used some filters known in the literature as Medium, Wiener, Frost, Lee and Wavelet (Coiflets and Fejer-Korovkin), to perform the mammographic image processing from 4 datasets, whose images were lauded as BI-RADS of category density 4. The performance of the filters was quantitatively evaluated using the following image quality parameters: Signal to Noise-Ratio (SNR), Peak Signal to Noise-Ratio (PSNR), Structural Similarity Index (SSIM) and Measure of Enhancement (EME). After processing with these filters, a new methodology was proposed in which a combination was made between the filters that obtained the best results. The proposed methodology proved to be effective for some groups of images, mainly considering the contrast metric, EME, where the combinations, in addition to showing that there is an increase of contrast in some cases, in the most of these maintain the EME values in relation to image, indicating that with the proposed methodology there is less blurring in the noise smoothing process. However, most wavelets and the wiener filter resulted in higher values of PSNR and SNR.