Redes neuronales convulucionales para mejorar la resolución de imágenes medicas

Abstract

Abstract: Mammography is one of the most commonly used techniques for detecting breast cancer. However, these images may present low quality in their resolution and contrast due to the noise produced by the equipment during the taking of the image. The main objective of this work is to improve the quality and super-resolution of mammography images through two algorithms based on convolutional neural networks. For the methodology (i) the CBIS-DDSM database is downloaded, (ii) the regions of interest (RoIs) are manually extracted. (iii) The data (RoIs) is increased, (iv) the super resolution of the RoIs is improved by means of the EDSR and RDN algorithms. (v) Finally, the quality of the new images generated is evaluated using the PSNR and SSIM evaluation metrics. The main results indicate that the RDN algorithm obtained an average of PSNR=32.68 dB and SSIM=0.82, and EDSR an average of PSNR=39.05 dB and SSIM=0.90 respectively. In conclusion, the experimental results demonstrate that the EDSR algorithm produces visibly clearer images with acceptable quality within the established range (38dB to 40dB)