Koen Van Leemput

The Multimodal Brain Tumor Image Segmentation Benchmark (BRATS)

In this paper we report the set-up and results of the Multimodal Brain Tumor Image Segmentation Benchmark (BRATS) organized in conjunction with the MICCAI 2012 and 2013 conferences. Twenty state-of-the-art tumor segmentation algorithms were applied to a set of 65 multi-contrast MR scans of low- and high-grade glioma patients - manually annotated by up to four raters - and to 65 comparable scans generated using tumor image simulation software. Quantitative evaluations revealed considerable disagreement between the human raters in segmenting various tumor sub-regions (Dice scores in the range 74%-85%), illustrating the difficulty of this task. We found that different algorithms worked best for different sub-regions (reaching performance comparable to human inter-rater variability), but that no single algorithm ranked in the top for all sub-regions simultaneously. Fusing several good algorithms using a hierarchical majority vote yielded segmentations that consistently ranked above all individual algorithms, indicating remaining opportunities for further methodological improvements. The BRATS image data and manual annotations continue to be publicly available through an online evaluation system as an ongoing benchmarking resource.

The Multimodal Brain TumorImage Segmentation Benchmark (BRATS)

In this paper we report the set-up and results of the Multimodal Brain Tumor Image Segmentation Benchmark (BRATS) organized in conjunction with the MICCAI 2012 and 2013 conferences. Twenty state-of-the-art tumor segmentation algorithms were applied to a set of 65 multi-contrast MR scans of low- and high-grade glioma patients - manually annotated by up to four raters - and to 65 comparable scans generated using tumor image simulation software. Quantitative evaluations revealed considerable disagreement between the human raters in segmenting various tumor sub-regions (Dice scores in the range 74%-85%), illustrating the difficulty of this task. We found that different algorithms worked best for different sub-regions (reaching performance comparable to human inter-rater variability), but that no single algorithm ranked in the top for all sub-regions simultaneously. Fusing several good algorithms using a hierarchical majority vote yielded segmentations that consistently ranked above all individual algorithms, indicating remaining opportunities for further methodological improvements. The BRATS image data and manual annotations continue to be publicly available through an online evaluation system as an ongoing benchmarking resource.

Automated segmentation of hippocampal subfields from ultra-high resolution in vivo MRI

Recent developments in MRI data acquisition technology are starting to yield images that show anatomical features of the hippocampal formation at an unprecedented level of detail, providing the basis for hippocampal subfield measurement. However, a fundamental bottleneck in MRI studies of the hippocampus at the subfield level is that they currently depend on manual segmentation, a laborious process that severely limits the amount of data that can be analyzed. In this article, we present a computational method for segmenting the hippocampal subfields in ultra‐high resolution MRI data in a fully automated fashion. Using Bayesian inference, we use a statistical model of image formation around the hippocampal area to obtain automated segmentations. We validate the proposed technique by comparing its segmentations to corresponding manual delineations in ultra‐high resolution MRI scans of 10 individuals, and show that automated volume measurements of the larger subfields correlate well with manual volume estimates. Unlike manual segmentations, our automated technique is fully reproducible, and fast enough to enable routine analysis of the hippocampal subfields in large imaging studies.

Automatic Brain Tumor Segmentation by Subject Specific Modification of Atlas Priors

RATIONALE AND OBJECTIVES Manual segmentation of brain tumors from magnetic resonance images is a challenging and time-consuming task. An automated system has been developed for brain tumor segmentation that will provide objective, reproducible segmentations that are close to the manual results. Additionally, the method segments white matter, grey matter, cerebrospinal fluid, and edema. The segmentation of pathology and healthy structures is crucial for surgical planning and intervention. MATERIALS AND METHODS The method performs the segmentation of a registered set of magnetic resonance images using an expectation-maximization scheme. The segmentation is guided by a spatial probabilistic atlas that contains expert prior knowledge about brain structures. This atlas is modified with the subject-specific brain tumor prior that is computed based on contrast enhancement. RESULTS Five cases with different types of tumors are selected for evaluation. The results obtained from the automatic segmentation program are compared with results from manual and semi-automated methods. The automated method yields results that have surface distances at roughly 1-4 mm compared with the manual results. CONCLUSION The automated method can be applied to different types of tumors. Although its performance is below that of the semi-automated method, it has the advantage of requiring no user supervision.

A computational atlas of the hippocampal formation using ex vivo, ultra-high resolution MRI: Application to adaptive segmentation of in vivo MRI.

Automated analysis of MRI data of the subregions of the hippocampus requires computational atlases built at a higher resolution than those that are typically used in current neuroimaging studies. Here we describe the construction of a statistical atlas of the hippocampal formation at the subregion level using ultra-high resolution, ex vivo MRI. Fifteen autopsy samples were scanned at 0.13 mm isotropic resolution (on average) using customized hardware. The images were manually segmented into 13 different hippocampal substructures using a protocol specifically designed for this study; precise delineations were made possible by the extraordinary resolution of the scans. In addition to the subregions, manual annotations for neighboring structures (e.g., amygdala, cortex) were obtained from a separate dataset of in vivo, T1-weighted MRI scans of the whole brain (1mm resolution). The manual labels from the in vivo and ex vivo data were combined into a single computational atlas of the hippocampal formation with a novel atlas building algorithm based on Bayesian inference. The resulting atlas can be used to automatically segment the hippocampal subregions in structural MRI images, using an algorithm that can analyze multimodal data and adapt to variations in MRI contrast due to differences in acquisition hardware or pulse sequences. The applicability of the atlas, which we are releasing as part of FreeSurfer (version 6.0), is demonstrated with experiments on three different publicly available datasets with different types of MRI contrast. The results show that the atlas and companion segmentation method: 1) can segment T1 and T2 images, as well as their combination, 2) replicate findings on mild cognitive impairment based on high-resolution T2 data, and 3) can discriminate between Alzheimers disease subjects and elderly controls with 88% accuracy in standard resolution (1mm) T1 data, significantly outperforming the atlas in FreeSurfer version 5.3 (86% accuracy) and classification based on whole hippocampal volume (82% accuracy).

Quantitative comparison of 21 protocols for labeling hippocampal subfields and parahippocampal subregions in in vivo MRI: Towards a harmonized segmentation protocol

OBJECTIVE An increasing number of human in vivo magnetic resonance imaging (MRI) studies have focused on examining the structure and function of the subfields of the hippocampal formation (the dentate gyrus, CA fields 1-3, and the subiculum) and subregions of the parahippocampal gyrus (entorhinal, perirhinal, and parahippocampal cortices). The ability to interpret the results of such studies and to relate them to each other would be improved if a common standard existed for labeling hippocampal subfields and parahippocampal subregions. Currently, research groups label different subsets of structures and use different rules, landmarks, and cues to define their anatomical extents. This paper characterizes, both qualitatively and quantitatively, the variability in the existing manual segmentation protocols for labeling hippocampal and parahippocampal substructures in MRI, with the goal of guiding subsequent work on developing a harmonized substructure segmentation protocol. METHOD MRI scans of a single healthy adult human subject were acquired both at 3 T and 7 T. Representatives from 21 research groups applied their respective manual segmentation protocols to the MRI modalities of their choice. The resulting set of 21 segmentations was analyzed in a common anatomical space to quantify similarity and identify areas of agreement. RESULTS The differences between the 21 protocols include the region within which segmentation is performed, the set of anatomical labels used, and the extents of specific anatomical labels. The greatest overall disagreement among the protocols is at the CA1/subiculum boundary, and disagreement across all structures is greatest in the anterior portion of the hippocampal formation relative to the body and tail. CONCLUSIONS The combined examination of the 21 protocols in the same dataset suggests possible strategies towards developing a harmonized subfield segmentation protocol and facilitates comparison between published studies.

Automatic Brain and Tumor Segmentation

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 subjects information about tumor location obtained from subtraction of post- and pre-contrast MRI. The new method handles various types of pathology, 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.

Predicting the location of entorhinal cortex from MRI

Entorhinal cortex (EC) is a medial temporal lobe area critical to memory formation and spatial navigation that is among the earliest parts of the brain affected by Alzheimers disease (AD). Accurate localization of EC would thus greatly facilitate early detection and diagnosis of AD. In this study, we used ultra-high resolution ex vivo MRI to directly visualize the architectonic features that define EC rostrocaudally and mediolaterally, then applied surface-based registration techniques to quantify the variability of EC with respect to cortical geometry, and made predictions of its location on in vivo scans. The results indicate that EC can be localized quite accurately based on cortical folding patterns, within 3 mm in vivo, a significant step forward in our ability to detect the earliest effects of AD when clinical intervention is most likely to be effective.

Segmentation of image ensembles via latent atlases.

Spatial priors, such as probabilistic atlases, play an important role in MRI segmentation. However, the availability of comprehensive, reliable and suitable manual segmentations for atlas construction is limited. We therefore propose a method for joint segmentation of corresponding regions of interest in a collection of aligned images that does not require labeled training data. Instead, a latent atlas, initialized by at most a single manual segmentation, is inferred from the evolving segmentations of the ensemble. The algorithm is based on probabilistic principles but is solved using partial differential equations (PDEs) and energy minimization criteria. We evaluate the method on two datasets, segmenting subcortical and cortical structures in a multi-subject study and extracting brain tumors in a single-subject multi-modal longitudinal experiment. We compare the segmentation results to manual segmentations, when those exist, and to the results of a state-of-the-art atlas-based segmentation method. The quality of the results supports the latent atlas as a promising alternative when existing atlases are not compatible with the images to be segmented.

Predicting the location of human perirhinal cortex, Brodmann's area 35, from MRI.

The perirhinal cortex (Brodmanns area 35) is a multimodal area that is important for normal memory function. Specifically, perirhinal cortex is involved in the detection of novel objects and manifests neurofibrillary tangles in Alzheimers disease very early in disease progression. We scanned ex vivo brain hemispheres at standard resolution (1 mm × 1 mm × 1 mm) to construct pial/white matter surfaces in FreeSurfer and scanned again at high resolution (120 μm × 120 μm × 120 μm) to determine cortical architectural boundaries. After labeling perirhinal area 35 in the high resolution images, we mapped the high resolution labels to the surface models to localize area 35 in fourteen cases. We validated the area boundaries determined using histological Nissl staining. To test the accuracy of the probabilistic mapping, we measured the Hausdorff distance between the predicted and true labels and found that the median Hausdorff distance was 4.0mm for the left hemispheres (n=7) and 3.2mm for the right hemispheres (n=7) across subjects. To show the utility of perirhinal localization, we mapped our labels to a subset of the Alzheimers Disease Neuroimaging Initiative dataset and found decreased cortical thickness measures in mild cognitive impairment and Alzheimers disease compared to controls in the predicted perirhinal area 35. Our ex vivo probabilistic mapping of the perirhinal cortex provides histologically validated, automated and accurate labeling of architectonic regions in the medial temporal lobe, and facilitates the analysis of atrophic changes in a large dataset for earlier detection and diagnosis.