Abstract. A framework is proposed for the segmentation of brain tumors from MRI. Instead of training on pathology, the proposed method trains exclusively on healthy tissue. The algorithm attempts to recognize deviations from normalcy in order to compute a fitness map over the image associated with the presence of pathology. The resulting fitness map may then be used by conventional image segmentation techniques for honing in on boundary delineation. Such an approach is applicable to structures that are too irregular, in both shape and texture, to permit construction of comprehensive training sets. The technique is an extension of EM segmentation that considers information on five layers: voxel intensities, neighborhood coherence, intra-structure properties, inter-structure relationships, and user input. Information flows between the layers via multi-level Markov random fields and Bayesian classification. A simple instantiation of the framework has been implemented to perform preliminary experiments on synthetic and MRI data. 1

Combining image segmentation based on statistical classification with a geometric prior has been shown to significantly increase robustness and reproducibility. Using a probabilistic geometric model of sought structures and image registration serves both initialization of probability density functions and definition of spatial constraints. A strong spatial prior, however, prevents segmentation of structures that are not part of the model. In practical applications, we encounter either the presentation of new objects that cannot be modeled with a spatial prior or regional intensity changes of existing structures not explained by the model. Our driving application is the segmentation of brain tissue and tumors from three-dimensional magnetic resonance imaging (MRI). Our goal is a high-quality segmentation of healthy tissue and a precise delineation of tumor boundaries. We present an extension to an existing expectation maximization (EM) segmentation algorithm that modifies a probabilistic brain atlas with an individual subject’s information about tumor location obtained from subtraction of post- and pre-contrast MRI. The new method handles various types of pathology, spaceoccupying mass tumors and infiltrating changes like edema. Preliminary results on five cases presenting tumor types with very different characteristics demonstrate the potential of the new technique for clinical routine use for planning and monitoring in neurosurgery, radiation oncology, and radiology. I.

In this work, we present a unique set of 3D MRI brain data that is appropriate for testing the intra and inter-site variability of image analysis methods. A single subject was scanned two times within a 24 hour time window each at five di#erent MR sites over a period of six weeks using GE and Phillips 1.5 T scanners. The imaging protocol included T1 weighted, Proton Density and T2 weighted images. We applied three quantitative image analysis methods and analyzed their results via the coe#cients of variability (COV) and the intra correlation coe#cient. The tested methods include two multi-channel tissue segmentation techniques based on an anatomically guided manual seeding and an atlas-based seeding. The third tested method was a single-channel semi-automatic segmentation of the hippocampus. The results show that the outcome of image analysis methods varies significantly for images from di#erent sites and scanners. With the exception of total brain volume, which shows consistent low variability across all images, the COV's were clearly larger between sites than within sites. Also, the COV's between sites with di#erent scanner types are slightly larger than between sites with the same scanner type. The presented existence of a significant inter-site variability requires adaptations in image methods to produce repeatable measurements. This is especially of importance in multi-site clinical research.

Abstract. We propose an extended graph-shifts algorithm for image segmentation and labeling. This algorithm performs energy minimization by manipulating a dynamic hierarchical representation of the image. It consists of a set of moves occurring at different levels of the hierarchy where the types of move, and the level of the hierarchy, are chosen automatically so as to maximally decrease the energy. Extended graph-shifts can be applied to a broad range of problems in medical imaging. In this paper, we apply extended graph-shifts to the detection of pathological brain structures: (i) segmentation of brain tumors, and (ii) detection of multiple sclerosis lesions. The energy terms in these tasks are learned from training data by statistical learning algorithms. We demonstrate accurate results, precision and recall in the order of 93%, and also show that the algorithm is computationally efficient, segmenting a full 3D volume in about one minute. 1

We present a method to detect intensity changes in longitudinal volumetric MRI data from patients with multiple sclerosis (MS).

The literature about partial volume (PV) segmentation of MR images is rather limited, and a general methodology for robustly classifying images with severe partial voluming that works well in all cases, remains an open issue. In this paper, we present a statistical framework for PV segmentation that contains and extends existing techniques. We think of a partial volumed image as a downsampled version of a fictive higher-resolution image that does not contain partial voluming, and we estimate the model parameters of this underlying image using an Expectation-Maximization algorithm. This leads to an iterative approach that interleaves a statistical classification of the image voxels using spatial information and an according update of the model parameters. We demonstrate on simulated data that the use of appropriate spatial prior knowledge, in casu a Markov random field model, not only improves the classifications, but is often indispensable for robust parameter estimation as well. We also present results on 2-D slices of real high-resolution MR images of the brain, and conclude that general robust segmentation of lower-resolution images requires development of spatial models that accurately describe the shape of the brain. 1

Abstract. A new algorithm is presented for the automatic segmentation of Multiple Sclerosis (MS) lesions in 3D MR images. It builds on the discriminative random decision forest framework to provide a voxel-wise probabilistic classification of the volume. Our method uses multi-channel MR intensities (T1, T2, Flair), spatial prior and long-range comparisons with 3D regions to discriminate lesions. A symmetry feature is introduced accounting for the fact that some MS lesions tend to develop in an asymmetric way. Quantitative evaluation of the data is carried out on publicly available labeled cases from the MS Lesion Segmentation Challenge 2008 dataset and demonstrates improved results over the state of the art. 1

Abstract — Combining image segmentation based on statistical classification with a geometric prior has been shown to significantly increase robustness and reproducibility. Using a probabilistic geometric model of sought structures and image registration serves both initialization of probability density functions and definition of spatial constraints. A strong spatial prior, however, prevents segmentation of structures that are not part of the model. In practical applications, we encounter either the presentation of new objects that cannot be modelled with a spatial prior or regional intensity changes of existing structures. Our driving application is the segmentation of brain tissue and tumors from three-dimensional magnetic resonance imaging (MRI). We aim at both obtaining a high-quality segmentation of healthy tissue and a precise delineation of tumor boundaries. We present an extension to an existing expectation maximization segmentation (EM) algorithm that modifies a probabilistic brain atlas with individual subject’s information about tumor location. This information is obtained from subtraction of post- and pre-contrast MRI and calculation of a posterior probability map for tumor. The new method handles both phenomena, space-occupying mass tumors and infiltrating changes like edema. Preliminary results on five cases presenting tumor types with very different characteristics demonstrate the potential of the new technique for clinical routine use for planning and monitoring in neurosurgery, radiation oncology, and radiology. I.

Abstract. Feature detection on MR images has largely relied on intensity classification and gradient-based magnitudes. In this paper, we propose the use of phase congruency as a more robust detection method, as it is based on a multiscale intensity-invariant measure. We show the application of phase congruency for the detection of cortical sulci from T2 weighted MRI. Sulci represent important landmarks in the structural analysis of the brain, as their location and orientation provide valuable information for diagnosis and surgical planning. Results show that phase congruency outperforms previous techniques, even in the presence of intensity bias fields due to magnetic field inhomogeneity. 1

This paper extends a conventional discounting scheme commonly used with the Dempster-Shafer evidential reasoning to deal with conflict. The extended discounting scheme is able to augment, discount, and oppose existing evidence structures when discounting factors take values in different ranges. To show its effectiveness, the scheme is employed for detecting multiple sclerosis (MS) lesions based on multi-modality MR images. The approach is fully automated and unsupervised. Experimental results have demonstrated that in addition to the superior segmentation accuracies of brain tissues, good MS detection performances have been obtained (MS detection accuracy 90.28%, similarity index 84.19%, and sensitivity 78.68% on average).