Brain tumor segmentation (BTS) using magnetic resonance imaging (MRI) plays a crucial role in the diagnosis, treatment, and research of brain tumors. With advancements in computer vision technology, numerous effective BTS models have been developed to tackle various challenges. However, most existing models overlook the incorporation of medical prior knowledge to guide the learning of intrinsic features beyond the label data. In this paper, we present PTransBTS, a prior-integrated brain tumor segmentation model that effectively leverages both the relationships between imaging modalities and lesion regions, as well as tumor shape priors. To address two key challenges in brain tumor segmentation—localization and precise delineation of fuzzy tumor boundaries—we introduce a Conv-head Global Attention (CHGA) block that combines parallel convolutional and Transformer-based architectures. Instead of directly merging all modalities, we propose a universal Brain Tumor Segmentation Stem (BTSS), which individually extracts features from each modality and applies weights for transmission to distinct decoder branches for segmenting different tumor subregions. To incorporate shape priors, we design a Learnable Dynamic Prior (LDP) block utilizing cross-attention mechanisms.