Traumatic brain injury (TBI) is one of the leading causes of disability and mortality worldwide, since a large proportion of the population is exposed to this kind of injury due to falls or traffic accidents. TBI can lead to short and long-term consequences, affecting the physical, mental, and functional health of individuals. These sequelae involve social, emotional, and professional challenges, directly impacting quality of life. TBI diagnosis is based on the study of injuries on Magnetic resonance imaging brain images; however, the accurate segmentation of TBI-related injuries remains challenging due to their variability in size, shape, location, and imaging characteristics, which limits the accuracy of diagnosis. This work proposes a strategy for the automatic segmentation of injuries associated with TBI based on nnU-Net, including an evaluation of different preprocessing techniques and fine-tuning to analyze their effect on the model’s performance. The results show that nnU-Net combined with KDE normalization and fine-tuning achieves the best performance (DSC = 0.56). Besides, a framework integrating the entire segmentation pipeline was implemented, providing a tool to support clinical practice and future research related to TBI.