Breast cancer is the type of cancer that develops from breast tissue. It is the most common cancer in women. Early detection of breast tumor is crucial in the treatment process. Magnetic Resonance Imaging (MRI) is a valuable tool for identifying and monitoring cancerous breast tumors and interpreting suspicious regions, because MRIs have excellent soft tissue imaging capability. However, this requires an experienced radiologist to analyze and interpret the data. On the other hand, image segmentation can help radiologists and doctors in the diagnosis of the disease and in the planning of its treatment. In this paper, we propose a new approach based on alliances in graphs for image segmentation, called GA2IS, which we apply to breast MRIs to extract existing tumors. The proposed approach has been tested and evaluated on the private dataset CMH-LIMED and has been compared with several concurrent methods in the literature. The obtained results, by considering several comparison metrics, have been in favor of the proposed GA2IS approach.