Redes neuronales para la segmentación de imágenes médicas

Abstract

Abstract: Breast cancer is one of the leading causes of death in women. The mammography technique is one of the most used for early detection of breast cancer, allowing an early diagnosis of breast lesions. However, manual reading implies an intense workload for the radiologist. For this reason, assisted computational systems (CAD) based on deep learning have been designed to help the doctor make decisions. In this work we focus on the use of convolutional networks to segment regions of interest (ROI) such as masses, tumors, microcalcifications and lesions in mammography images. For this, it is proposed 1) Use the CBIS-DDSM database, 2) Obtain the ROI manually, 3) Segmentation of the ROI using the U-net and SegNet algorithms, 4) Evaluation of the performance of these algorithms using statistical metrics. Finally, the results indicate that U-net exceeds the accuracy percentage 88% with respect to that of SegNet 82%. In conclusion, the results indicate that the U-net model is more robust and accurate for segmenting regions in the breast.