We introduce Atropos, an ITK-based multivariate n-class open source segmentation algorithm distributed with ANTs (http://www.picsl.upenn.edu/ANTs). The Bayesian formulation of the segmentation problem is solved using the Expectation Maximization (EM) algorithm with the modeling of the class intensities based on either parametric or non-parametric finite mixtures. Atropos is capable of incorporating spatial prior probability maps (sparse), prior label maps and/or Markov Random Field (MRF) modeling. Atropos has also been efficiently implemented to handle large quantities of possible labelings (in the experimental section, we use up to 69 classes) with a minimal memory footprint. This work describes the technical and implementation aspects of Atropos and evaluates its performance on two different ground-truth datasets. First, we use the BrainWeb dataset from Montreal Neurological Institute to evaluate three-tissue segmentation performance via (1) K-means segmentation without use of template data; (2) MRF segmentation with initialization by prior probability maps derived from a group template; (3) Prior-based segmentation with use of spatial prior probability maps derived from a group template. We also evaluate Atropos performance by using spatial priors to drive a 69-class EM segmentation problem derived from the Hammers atlas from University College London. These evaluation studies, combined with illustrative examples that exercise Atropos options, demonstrate both performance and wide applicability of this new platform-independent open source segmentation tool.

An Open Source Multivariate Framework for n-Tissue Segmentation with Evaluation on Public Data

The variation between images obtained with different scanners or different imaging protocols presents a major challenge in automatic segmentation of biomedical images. This variation especially hampers the application of otherwise successful supervised-learning techniques which, in order to perform well, often require a large amount of labeled training data that is exactly representative of the target data. We therefore propose to use transfer learning for image segmentation. Transfer-learning techniques can cope with differences in distributions between training and target data, and therefore may improve performance over supervised learning for segmentation across scanners and scan protocols. We present four transfer classifiers that can train a classification scheme with only a small amount of representative training data, in addition to a larger amount of other training data with slightly different characteristics. The performance of the four transfer classifiers was compared to that of standard supervised classification on two magnetic resonance imaging brain-segmentation tasks with multi-site data: white matter, gray matter, and cerebrospinal fluid segmentation; and white-matter-/MS-lesion segmentation. The experiments showed that when there is only a small amount of representative training data available, transfer learning can greatly outperform common supervised-learning approaches, minimizing classification errors by up to 60%.

/pdf/transfer-learning-improves-supervised-image-segmentation-2z4fu4ou7c.pdf

Transfer Learning Improves Supervised Image Segmentation Across Imaging Protocols

Segmentation of multiple sclerosis lesions in brain MRI: A review of automated approaches

We compare three widely used brain volumetry methods available in the software packages FSL, SPM5, and FreeSurfer and evaluate their performance using simulated and real MR brain data sets. We analyze the accuracy of gray and white matter volume measurements and their robustness against changes of image quality using the BrainWeb MRI database. These images are based on "gold-standard" reference brain templates. This allows us to assess between- (same data set, different method) and also within-segmenter (same method, variation of image quality) comparability, for both of which we find pronounced variations in segmentation results for gray and white matter volumes. The calculated volumes deviate up to >10% from the reference values for gray and white matter depending on method and image quality. Sensitivity is best for SPM5, volumetric accuracy for gray and white matter was similar in SPM5 and FSL and better than in FreeSurfer. FSL showed the highest stability for white ( 20% for the simulated data and 24% on average for the real data sets, whereas within-method performance analysis uncovered volume differences of up to >15%. Since the discrepancies between results reach the same order of magnitude as volume changes observed in disease, these effects limit the usability of the segmentation methods for following volume changes in individual patients over time and should be taken into account during the planning and analysis of brain volume studies.

/pdf/evaluation-of-automated-brain-mr-image-segmentation-and-32kta1pknq.pdf

Evaluation of automated brain MR image segmentation and volumetry methods.

The presence of lymphocytic infiltration (LI) has been correlated with nodal metastasis and tumor recurrence in HER2+ breast cancer (BC). The ability to automatically detect and quantify extent of LI on histopathology imagery could potentially result in the development of an image based prognostic tool for human epidermal growth factor receptor-2 (HER2+) BC patients. Lymphocyte segmentation in hematoxylin and eosin (H&E) stained BC histopathology images is complicated by the similarity in appearance between lymphocyte nuclei and other structures (e.g., cancer nuclei) in the image. Additional challenges include biological variability, histological artifacts, and high prevalence of overlapping objects. Although active contours are widely employed in image segmentation, they are limited in their ability to segment overlapping objects and are sensitive to initialization. In this paper, we present a new segmentation scheme, expectation-maximization (EM) driven geodesic active contour with overlap resolution (EMaGACOR), which we apply to automatically detecting and segmenting lymphocytes on HER2+ BC histopathology images. EMaGACOR utilizes the expectation-maximization algorithm for automatically initializing a geodesic active contour (GAC) and includes a novel scheme based on heuristic splitting of contours via identification of high concavity points for resolving overlapping structures. EMaGACOR was evaluated on a total of 100 HER2+ breast biopsy histology images and was found to have a detection sensitivity of over 86% and a positive predictive value of over 64%. By comparison, the EMaGAC model (without overlap resolution) and GAC model yielded corresponding detection sensitivities of 42% and 19%, respectively. Furthermore, EMaGACOR was able to correctly resolve over 90% of overlaps between intersecting lymphocytes. Hausdorff distance (HD) and mean absolute distance (MAD) for EMaGACOR were found to be 2.1 and 0.9 pixels, respectively, and significantly better compared to the corresponding performance of the EMaGAC and GAC models. EMaGACOR is an efficient, robust, reproducible, and accurate segmentation technique that could potentially be applied to other biomedical image analysis problems.

Expectation–Maximization-Driven Geodesic Active Contour With Overlap Resolution (EMaGACOR): Application to Lymphocyte Segmentation on Breast Cancer Histopathology

An automated algorithm for tissue segmentation of noisy, low-contrast magnetic resonance (MR) images of the brain is presented. A mixture model composed of a large number of Gaussians is used to represent the brain image. Each tissue is represented by a large number of Gaussian components to capture the complex tissue spatial layout. The intensity of a tissue is considered a global feature and is incorporated into the model through tying of all the related Gaussian parameters. The expectation-maximization (EM) algorithm is utilized to learn the parameter-tied, constrained Gaussian mixture model. An elaborate initialization scheme is suggested to link the set of Gaussians per tissue type, such that each Gaussian in the set has similar intensity characteristics with minimal overlapping spatial supports. Segmentation of the brain image is achieved by the affiliation of each voxel to the component of the model that maximized the a posteriori probability. The presented algorithm is used to segment three-dimensional, T1-weighted, simulated and real MR images of the brain into three different tissues, under varying noise conditions. Results are compared with state-of-the-art algorithms in the literature. The algorithm does not use an atlas for initialization or parameter learning. Registration processes are therefore not required and the applicability of the framework can be extended to diseased brains and neonatal brains

/pdf/constrained-gaussian-mixture-model-framework-for-automatic-3rpizs53wx.pdf

A. Ruf

Papers

Constrained Gaussian mixture model framework for automatic segmentation of MR brain images