Automated peritumoral edema segmentation in preoperative brain MRI scans

Peritumoral brain edema is a frequently encountered phenomenon that strongly contributes to neurological signs and symptoms. This paper presents a fully automated procedure based on Support Vector Machine for the accurate and efficient edema delineation in brain MRI imagery. The segmentation procedure is conceived fully automatic and specifically oriented to non-volumetric data to facilitate the insertion in clinical practice. The major efforts in building a robust classifier were concentrated in the extraction and selection of a set of significant features from multispectral brain MR images. Preliminary experimental results obtained by processing in-house collected data, are encouraging creating the premises for an effective use of the automated segmentation procedure in brain tumor clinical studies. To see if the proposed supervised method can be considered an alternative to other contemporary approaches, the results obtained were compared with those obtained by an unsupervised region-based segmentation procedure based on Graph Cut.