A neural tool for breast cancer detection and classification in MRI

Breast cancer is a major problem for the healthcare systems of industrialized countries. It's clear that an improvement of early diagnostic techniques would be very important for women's quality of life. Actually contrast-enhanced magnetic resonance of the breast is the most attractive alternative to standard mammography. Due to its rich semeiotics and to the number of images that have to be analyzed, the manual inspection of a contrast-enhanced study is a long and error-prone process that produces subjective results often as good as the clinician's experience. The major problem with breast cancer is the significant overlap between features of benign and malignant tissues: apart from simple cases, there is no standard and universally accepted methodology for tissue classification. In this paper we present an automatic tool for breast image analysis. The tool is organized as a pipeline of stages, the most important one being the neural module. This module uses advanced neural architectures to exploit important statistical relationships between the features of the different tissue types. As a future goal, we wish to establish, within the built framework, the exact relationship between the different features extracted from relevant tissues.