Data driven brain tumor segmentation in MRI using probabilistic reasoning over space and time

Abstract

A semi-automated method for brain tumor segmentation and volume tracking has been developed. This method uses a pipeline approach to process MRI images. The pipeline process involves the following steps: 1) automatic alignment of initial and subsequent MRI scans for a given patient, 2) automatic de-skulling of all brain images, 3) automatic segmentation of the brain tumors using probabilistic reasoning over space and time with a semi-automatic correction of segmentation results on the first time point only and, 4) brain tumor tracking, providing a report of tumor volume change. To validate the procedure, we evaluated contrast enhanced MRI images from five brain tumor patients, each scanned at three times, several months apart. This data was processed and estimated tumor volume results show good agreement with manual tracing of 3D lesions over time. © Springer-Verlag Berlin Heidelberg 2004.