Bayesian computer vision methods for improved tumor localization and delineation

The authors are measuring the effectiveness of Bayesian computer-vision methods that are designed to automatically delineate tissue and tumor boundaries by incorporating 3-D information from a multispectral MRI (magnetic resonance imaging) data set. This data set can be directly exploited for tissue characterization and analysis by simply designing the Bayesian computer vision model to be robust to the complex nature of the multispectral imaging signal. Using this Bayes model one can then begin to form boundaries around the tumor and organs of interest by adopting Bayesian priors that assure boundary connectivity and structural integrity. Once boundaries are formed that are consistent with the observed MRI data and the Bayesian prior constraints, this boundary information can be combined with the 3-D volumetric image data to construct a highly informative 3-D display to be used for diagnosis, for treatment planning, or in the operating room as an aid to guiding surgery, particularly in the case of stereotactic surgery.<>