A brain tumor segmentation framework based on outlier detection q Marcel Prastawa a, * , Elizabeth Bullitt c , Sean Ho a , Guido Gerig a,b a Department of Computer Science, University of North Carolina, CB #3175, Sitterson Hall, Chapel Hill, NC 27599, USA b Department of Psychiatry, University of North Carolina, Chapel Hill, NC 27599, USA c Department of Surgery, University of North Carolina, Chapel Hill, NC 27599, USA Available online 17 July 2004 Abstract This paper describes a framework for automatic brain tumor segmentation from MR images. The detection of edema is done simultaneously with tumor segmentation, as the knowledge of the extent of edema is important for diagnosis, planning, and treatment. Whereas many other tumor segmentation methods rely on the intensity enhancement produced by the gadolinium contrast agent in the T1-weighted image, the method proposed here does not require contrast enhanced image channels. The only required input for the segmentation procedure is the T2 MR image channel, but it can make use of any additional non-enhanced image channels for improved tissue segmentation. The segmentation framework is composed of three stages. First, we detect ab- normal regions using a registered brain atlas as a model for healthy brains. We then make use of the robust estimates of the location and dispersion of the normal brain tissue intensity clusters to determine the intensity properties of the different tissue types. In the second stage, we determine from the T2 image intensities whether edema appears together with tumor in the abnormal regions. Finally, we apply geometric and spatial constraints to the detected tumor and edema regions. The segmentation procedure has been applied to three real datasets, representing different tumor shapes, locations, sizes, image intensities, and enhancement. Ó 2004 Published by Elsevier B.V. Keywords: Automatic brain segmentation; Brain tumor segmentation; Level-set evolution; Outlier detection; Robust estimation 1. Introduction Automatic brain tumor segmentation from MR im- ages is a difficult task that involves various disciplines covering pathology, MRI physics, radiologist’s percep- tion, and image analysis based on intensity and shape. There are many issues and challenges associated with brain tumor segmentation. Brain tumors may be of any size, may have a variety of shapes, may appear at any location, and may appear in different image intensities. Some tumors also deform other structures and appear together with edema that changes intensity properties of the nearby region. For many human experts, manual segmentation is a difficult and time consuming task, which makes an automated brain tumor segmentation method desirable. There are many possible applications of an automated method, it can be used for surgical plan- ning, treatment planning, and vascular analysis. It has been shown that blood vessels in the brain exhibit certain characteristics within pathological regions (Bullitt et al., 2003). An objective and reproducible segmentation pro- cedure coupled with vascular analysis would allow us to study the relation between pathologies and blood vessels and may function as a new diagnostic measure. The challenges associated with automatic brain tu- mor segmentation have given rise to many different approaches. Automated segmentation methods based on artificial intelligence techniques were proposed in (Clark et al., 1998; Fletcher-Heath et al., 2001). The two methods do not rely on intensity enhancements provided by the use of contrast agents. A particular limitation of the two methods is that the input images are restricted to the T1, T2, and PD MR image channels. Addition- ally, the methods require a training phase prior to seg- menting a set of images. Other methods are based on statistical pattern recognition techniques, for example the method proposed by Kaus et al. (1999). This method q Supported by NIH-NIBIB R01 EB000219 and NIH-HLB R01 HL69808. * Corresponding author. Tel.: +919-962-1836. E-mail address: prastawa@cs.unc.edu (M. Prastawa). 1361-8415/$ - see front matter Ó 2004 Published by Elsevier B.V. doi:10.1016/j.media.2004.06.007 Medical Image Analysis 8 (2004) 275–283 www.elsevier.com/locate/media