Aumento de imágenes de mama mediante técnicas de aprendizaje profundo

Abstract

Abstract: Ultrasound and mammography techniques are among the most used for early detection of breast cancer due to their low radial exposure to the patient and their low cost, allowing early diagnosis of breast lesions. Artificial intelligence techniques are currently being used to augment medical images, especially due to the challenge that obtaining them represents to train algorithms based on deep learning. In this paper we specifically focus on the use of GAN networks to increase breast regions of interest (ROI). To this end, the following objectives are proposed: 1) Implement two algorithms based on GAN networks for augmentation of breast images with ultrasound and mammography, 2) Evaluate and compare the performance of the algorithms using the KID evaluation metric. The results obtained show that the CycleGAN algorithm generated better quality images with a KID score of 0.20 compared to the WGAN-GP algorithm with a score of 0.60. In conclusion, the experimental results indicate that the CycleGAN model is more robust and accurate than WGAN-GP for enlargement of breast regions.