Locating Brain Tumors from MR Imagery Using Symmetry

Tumor/abnormality segmentation from magnetic resonance imagery (MRI) can play a significant role in cancer research and clinical practice. Although accurate tumor segmentation by radiologists is ideal, it is extremely tedious. Experience shows that for MRI database indexing purposes approximate segmentations can be adequate. In this paper, we propose a straightforward, real-time technique to find a bounding box around the brain abnormality in an MR image. Our algorithm exploits left-to-right symmetry of the brain structure. The proposed detection algorithm can play a useful role in indexing and storage of bulk MRI data, as well as provide an initial step or seed to assist algorithms designed to find accurate tumor boundaries.