Automated evaluation of hippocampus volume plays a crucial role in the analysis of various neurodegenerative conditions like Alzheimer’s Disease and Epilepsy. Examination of the hippocampus subfields assumes paramount importance as it can reveal early signs of brain abnormalities. However, delineating these subfields becomes extremely challenging due to their intricate nature and the requirement for manually annotated high-resolution magnetic resonance images. In this paper, we propose an innovative deep graph cut approach, boosted by shape information, for automatic segmentation of hippocampus subfields. A deep learned shape term is incorporated in the energy function of the graph cut. A modified \(\alpha -\beta \) swap technique, that leverages deep learning, is designed to improve the execution time of the proposed multi-class segmentation algorithm. We demonstrate the efficacy of our solution by outperforming a number of state-of-the-art methods on the publicly available Kulaga-Yoskovitz dataset.