Brain tumor segmentation is an important image processing step in diagnosis, treatment planning, and follow-up studies of Glioblastoma (GBM). However it is still a challenging task due to varying in size, shape, location, and image intensities within and around the tumor. In this paper, we propose a new brain tumor segmentation method for T1-weighted MR brain images based on an improved level set method using prior information as a constraint, called Prior Information Constrained Evolution (PICE). A new energy function in PICE incorporating the tumor intensity prior is designed to match brain tumor more accurately. The advantage of PICE has been illustrated by comparing with the traditional level set method in 3-D. In addition, we also illustrate that PICE can be easily applied to 4-D images, which facilitates follow-up studies of brain tumor treatments. Using longitudinal GBM data from five patients we showed the advantages of the proposed algorithm.