In recent years, the research of image processing is developed from traditional two dimensions (2D) to three dimensions (3D) or even more dimensions. However, the existing segmentation methods are mainly based on 2D image processing, and more effective 3D image segmentation methods are expected. An adaptive 3D image segmentation method based on vector quantization (VQ) that can effectively utilize the spatial information of the volume data of the 3D image is proposed. In the method, a preprocessing is conducted on the 3D image, including volume interpolation, dividing of the 3D image into small sub-cubic blocks (sub-cubes), and classification of the sub- cubes into two patterns, the edge pattern and non- edge pattern. The non- edge pattern sub- cubes are segmented by using the VQ technique and the edge pattern sub- cubes are classified in pixel based on the segmentation results of non-edge pattern sub-cubes. In order to determine the segmentation number adaptively, an optimal codebook searching algorithm is designed for the VQ approach. Experiments are conducted by using both the simulation samples and real human brain magnetic resonance imaging (MRI) images from the IBSR database and the effectiveness of the proposed method is validated by the experimental results. The experiments are also performed on the MRI images of the same patient in different treatment periods, which can provide the varied 3D information about the focus parts in different times that is valuable for clinical diagnosis in medical practice.