The detection, diagnosis, and treatment of prostate cancer depends on the correct determination of the prostate anatomy. In current practice, the prostate segmentation is performed manually by a radiologist, which is extremely time-consuming that demands experience and concentration. Therefore, this paper proposes an automatic method for prostate segmentation on 3D magnetic resonance imaging scans using a superpixel technique, phylogenetic indexes, and an optimized XGBoost algorithm. The proposed method has been evaluated on the Prostate 3T and PROMISE12 databases presenting a dice similarity coefficient of 84.48% and a volumetric similarity of 95.91%, demonstrating the high-performance potential of the proposed method.