Glioblastoma is a highly aggressive primary brain tumor with histopathological heterogeneity. Identifying relevant structures in histology images is crucial for diagnosis and treatment, yet manual interpretation is tedious. To address this, we developed a deep learning model for the classification of histopathological subregions in glioblastoma. In this work, we present one of the top-performing methods in the BraTS-Path 2025 Challenge, a competition aimed at developing computational pathology solutions for glioblastoma. We adapted a Vision Transformer pretrained on a large-scale histology dataset. To achieve a balance between efficiency and performance, we employed parameter-efficient fine-tuning with Low-Rank Adaptation. To address class imbalance and enhance generalization, we used stratified cross-validation, data augmentation, and ensemble inference. The results indicate that our approach performs robustly and can assist histopathological analysis in neuro-oncology. The code is available at http://github.com/oikosohn/brats25-path .