Elsevier

Information Sciences

Volume 186, Issue 1, 1 March 2012, Pages 164-185
Information Sciences

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

https://doi.org/10.1016/j.ins.2011.10.011Get rights and content

Abstract

Automatic segmentation of multiple sclerosis (MS) lesions in brain MRI has been widely investigated in recent years with the goal of helping MS diagnosis and patient follow-up. However, the performance of most of the algorithms still falls far below expert expectations. In this paper, we review the main approaches to automated MS lesion segmentation. The main features of the segmentation algorithms are analysed and the most recent important techniques are classified into different strategies according to their main principle, pointing out their strengths and weaknesses and suggesting new research directions. A qualitative and quantitative comparison of the results of the approaches analysed is also presented. Finally, possible future approaches to MS lesion segmentation are discussed.

Introduction

Multiple sclerosis (MS) is a chronic, persistent inflammatory-demyelinating and degenerative disease of the central nervous system (CNS), characterised pathologically by areas of inflammation, demyelination, axonal loss, and gliosis scattered throughout the CNS, often causing motor, sensorial, vision, coordination, deambulation, and cognitive impairment [28]. The two main clinical phenomena of prototypic MS are relapses and progression. Relapses are considered to be the clinical expression of acute focal or multifocal inflammatory demyelination, disseminated within the CNS. The remission of symptoms early in the disease is likely to be the result of remyelination, resolution of inflammation, and compensatory mechanisms such as the redistribution of axolemmal sodium channels and cortical plasticity. These recovery mechanisms are less effective after recurrent attacks.

Multiple sclerosis is the most frequent, non-traumatic, neurological disease capable of causing disability in young adults. It is relatively common in Europe, the United States, Canada, New Zealand and parts of Australia, but rare in Asia and the tropics and subtropics of all continents. Within regions having a temperate climate, the incidence and prevalence of MS increase with latitude – both north and south of the equator. Multiple sclerosis is between two and three-times more common in women than in men, but men have a tendency for later disease onset with a poorer prognosis. The incidence of MS is low in childhood, increases rapidly after the age of 18, reaches a peak between 25 and 35 and then slowly declines, becoming rare at 50 and older. It is estimated that there are between 1.3 and 2.5 million cases of MS in the world, with some 350,000 cases in Western Europe [35]. According to the latest epidemiological studies, the prevalence and incidence of MS has been increasing worldwide.

MS does not usually shorten life significantly, but there is a substantial impact on personal, social and work activities in patients and their families and its socioeconomic importance is second only to trauma in young adults. The etiology of MS is still unknown, but it appears most likely to be the result of interplay between as yet unidentified environmental factors and susceptible genes. Along with the demyelinating episodes, there may be damage to the exposed axons, leading to transection of the axons and retrograde neuronal degeneration. This process can be irreversible and is responsible for the accrual of disability that occurs as the disease progresses.

Conventional Magnetic Resonance Imaging (MRI) techniques [50], such as T2-weighted (T2-w) and gadolinium-enhanced T1-weighted (T1-w) sequences, are highly sensitive in detecting MS plaques and can provide quantitative assessment of inflammatory activity and lesion load. MRI-derived metrics have become the most important paraclinical tool for diagnosing MS, for understanding the natural history of the disease and for monitoring the efficacy of experimental treatments. Both acute and chronic MS plaques appear as focal high-signal intensity areas on T2-w sequences, reflecting their increased tissue water content. The increase in the signal indicates edema, inflammation, demyelination, reactive gliosis and/or axonal loss in proportions that differ from lesion to lesion. They are typically discrete and focal at the early stages of the disease, but become confluent as the disease progresses. The total T2 lesion volume of the brain increases by approximately 5–10% each year in the relapsing forms of MS [65]. Gadolinium-enhanced T1-w imaging is highly sensitive in detecting inflammatory activity. This technique detects disease activity 5 to 10 times more frequently than clinical evaluation of relapses, suggesting that most of the enhancing lesions (EL) are clinically silent. Longitudinal and cross-sectional MR studies have shown that the formation of new MS plaques is often associated with contrast enhancement, mainly in the acute and relapsing stages of the disease.

Approximately 10–20% of T2 hyperintense lesions (HL) are also visible on T1-w images as areas of low signal intensity compared with normal-appearing white matter (WM). These so-called T1 black holes (BH) have a different pathological substrate that depends, in part, on the lesion age. The hypointensity is present in up to 80% of recently formed lesions and probably represents marked edema, with or without myelin destruction or axonal loss. In most cases, the acute (or wet) black holes become isointense within a few months as inflammatory activity abates, edema resolves and reparative mechanisms like remyelination become active. Less than 40% evolve into persisting or chronic black holes [13]. Chronic black holes correlate pathologically with the most severe demyelination and axonal loss, indicating areas of irreversible tissue damage. T1-w sequences have a higher specificity than T2-w sequences for detecting lesions with irreversible tissue damage and may serve as surrogate markers of disability progression in clinical trials. Fig. 1 shows examples of MRI scans of a patient with MS lesions.

Atrophy of the brain and spinal cord is an important part of MS pathology and a clinically relevant component progression of the disease [17]. CNS atrophy, which involves both grey matter (GM) and white matter (WM), is a progressive phenomenon that worsens with increasing disease duration, and progresses at a rate of between 0.6% and 1.2% of brain loss per year. Quantitative measures of whole-brain atrophy can be acquired by automated or semi-automated methods that display this progressive loss of brain tissue bulk in vivo in a sensitive and reproducible manner.

In clinical trials as well as in every-day clinical practice, MRI scans are visually assessed for qualitative analysis and manually marked if a quantitative analysis is required. Quantitative analysis has become invaluable in the assessment of disease progression [75], [76] and the evaluation of therapies over the last 25 years [45], [23]. In fact, MRI metrics have become common primary endpoints in phase II immunomodulatory drug therapy trials [91]. Moreover, magnetisation transfer MRI, diffusion tensor MRI (DTI), proton MR spectroscopy, and functional MRI are nowadays also contributing to elucidate the mechanisms that underlie injury, repair, and functional adaptation in patients with MS [37], [64].

In quantitative analyses of focal lesions, in both cross-sectional and longitudinal studies, manual or semi-automated segmentations have been used to compute the total number of lesions and the total lesion volume. The manual delineation of MS lesions, however, is both challenging and time-consuming because of the large number of MRI slices required to compose three-dimensional information for each patient. Moreover, it is prone to intra-observer variability (the same study analysed by the same neuroradiologist at different times) and inter-observer variability (the same study analysed by different neuroradiologists), and requires strategies such as the STAPLE algorithm [107] to fuse different segmentation results into one. Another procedure used in clinical practice is scan-rescan reproducibility [68], which image processing practitioners often overlook when assessing the true reproducibility of the whole measurement process. Scan-rescan reproducibility consists of performing multiple scans of the same subject within a period of time and then assessing the reproducibility of both human observers and automated segmentations.

The development of fully automated MS segmentation methods, which can segment large amounts of MRI data and do not suffer from intra- and inter-observer variability, has become an active research field [95], [8]. Unfortunately, the results of these fully automated methods show less agreement with manually segmented scans than those obtained with segmentations by independent observers. Moreover, when evaluating MS segmentation methods, there are still no in vivo methods to obtain reliable ground truth data mainly because of the large intra- and inter-observer variability. It should be noted that disagreements between segmentations may have important influences on certain evaluation measures due to the small volume of the lesions.

In this paper, the state-of-the-art of strategies for automated MS lesion segmentation is reviewed with the aim of pointing out their strengths and weaknesses and suggest new research directions. Moreover, recent significant works in this field are described and the different techniques are classified according to the strategy used. In particular, a first classification between supervised and unsupervised segmentation strategies is established, dividing the supervised group into atlas-based approaches and those based on training by means of features extracted from manual segmentations. Moreover, the unsupervised group is further divided into those techniques that use tissue segmentation to obtain the lesions and those that use only the lesion properties for the segmentation. In addition to describing and classifying these approaches, a description of the algorithms used to segment the lesions as well as the features and the type of MR images used is also provided.

Various reviews of brain MRI segmentation have been presented in the past. For instance, Bezdek et al. [19] analysed 90 papers on MRI segmentation using pattern recognition techniques. The authors suggested dividing the algorithms into two categories: supervised methods (such as Bayes classifiers with labelled maximum likelihood estimators, the k-nearest neighbour rule (kNN), and artificial neural networks (ANN)) and unsupervised methods (i.e. Bayes classifiers with unlabelled maximum likelihood estimators or the fuzzy C-means (FCM) algorithms). In addition Clarke et al. [25] reviewed not only methods for MRI segmentation, but also general pre-processing algorithms, validation methods and registration between different MR images. However, these reviews were only related to soft brain tissue segmentation. More recently, Souplet et al. [93] presented a review of semi-automated and automated MS lesion segmentation approaches, analysing MS lesions, pre-processing steps and segmentation approaches. However, to the best of our knowledge, this paper is the first attempt to review the most relevant works in automated MS lesion segmentation that provides an evaluation of the experimental results.

The inability to compare evaluation results due to the use of different data sets and different evaluation measures has been a major obstacle to reviewing MS lesion segmentation methods. Ideally, methods should be applied to a common database and compared to a ground truth. This is however very difficult due to the lack of common public databases of real images along with their ground truth and the fact that only few methods are publicly available. Our contribution is though close to this idea. Here we will compare quantitatively the segmentation approaches accordingly to their reported results in the literature. Furthermore, the recent MS Lesion Segmentation Challenge [95] provided a common framework for MS lesion segmentation algorithms, allowing comparisons to be made between different approaches. In the results section of this paper, the most typical measures used for evaluating MS lesion segmentation results are described. Moreover, the works analysed are compared in a qualitative and quantitative way.

The rest of this paper is organised as follows. Section 2 reviews the pre-processing steps needed when automatically segmenting MS lesions. Section 3 shows the classification of the segmentation approaches and also reviews the features and algorithms used. In Section 4, the different image databases and measures used to evaluate the results are presented and used to compare the performances of the works analysed. Discussions are given in Section 5, whilst the paper finishes with some conclusions and suggestions for future work.

Section snippets

Pre-processing

The segmentation of MR brain images is difficult because of variable imaging parameters, overlapping intensities, noise, partial voluming, gradients, motion, echoes, blurred edges, normal anatomical variations and susceptibility artifacts [81]. Therefore, before applying any approach to MS lesion segmentation, there are generally two pre-processing steps that are carried out: first, the removal of those image artefacts and second, the removal of non-brain tissue, such as the skull, from the

MS lesion segmentation

In this section, the recent state-of-the-art of automated MS lesion segmentation is reviewed. Firstly, the main image features used as input for the different segmentation algorithms are analysed. Afterwards, a classification of the different strategies is proposed and the most significant works in this field are described. The approaches reviewed are summarised in Table 1, which offers a compact at-a-glance overview of these studies.

Results

MS lesion segmentation approaches are usually evaluated using different quantitative measures and both synthetic and real MRI volumes. The most common data sets used in the works analysed and the typical measures computed for the evaluation are described in this section. Finally, a comparison and discussion of the results presented by the different approaches, highlighting the most interesting aspects, is also presented.

Discussion

As seen in previous sections, the most widely-used feature in all segmentation methods is voxel intensity, which is commonly employed with a multi-channel approach. In addition, features based on modelling the voxel neighbourhood are also used in some approaches to introduce (local) spatial information to the algorithms. Regarding the modalities, T1-w images are widely used for the tissue segmentation and also for the black holes and enhanced lesion segmentation. T2-w and PD-w images are

Future trends

In this paper several strategies to perform the automated MS lesion segmentation have been reviewed. For instance, atlas-based segmentation is becoming a standard paradigm for exploiting spatial prior knowledge in MR brain image segmentation. Atlases provide helpful information about anatomy and its variability. Several works already showed that atlas selection, or a group of atlases, is crucial for improving the segmentation results. Thus, research interests are currently on the construction

Conclusions

This paper has reviewed the automated approaches for MS lesion segmentation, classifying them according to the strategy used. In addition, the results obtained by these approaches have been summarised and compared, reviewing also the most common data sets and evaluation measures used in this field. We observed that the automated segmentation of different MS lesion types in MRI is a challenging task due to heterogeneous intensity values among the different MR images (enhancing lesions, black

Acknowledgements

We would like to thank Dr. D. García-Lorenzo, Prof. C. Barillot, Prof. P. A. Narayana, and Dr. N.K. Subbanna, for providing the images of their results (Fig. 6). We would like to thank also the reviewers for their critical evaluation of the manuscript. This work has been supported by the Instituto de Salud Carlos III Grant PI09/91018, Grant VALTEC09-1-0025 from the Generalitat de Catalunya, and Grant CEM-Cat 2011 from the Fundació Esclerosi Múltiple. M. Cabezas holds a FI Grant 2011FI-B1 00167.

References (113)

  • R. Cárdenes et al.

    A multidimensional segmentation evaluation for medical image data

    Comput. Methods Prog. Biomed.

    (2009)
  • L.P. Clarke et al.

    MRI segmentation: methods and applications

    Magn. Reson. Imag.

    (1995)
  • S. Datta et al.

    Segmentation and quantification of black holes in multiple sclerosis

    NeuroImage

    (2006)
  • K.S. Fu et al.

    A survey on image segmentation

    Pattern Recog.

    (1981)
  • D. Goldberg-Zimring et al.

    Automated detection and characterization of multiple sclerosis lesions in brain MR images

    Magn. Reson. Imag.

    (1998)
  • A.S. Hadi et al.

    Maximum trimmed likelihood estimators: a unified approach, examples, and algorithms

    Comput. Stat. Data Anal.

    (1997)
  • S.W. Hartley et al.

    Analysis and validation of automated skull stripping tools: A validation study based on 296 MR images from the Honolulu Asia aging study

    NeuroImage

    (2006)
  • R. Khayati et al.

    Fully automatic segmentation of multiple sclerosis lesions in brain MR FLAIR images using adaptive mixtures method and Markov random field model

    Comput. Biol. Med.

    (2008)
  • M. Kuklisova-Murgasova et al.

    A dynamic 4D probabilistic atlas of the developing brain

    NeuroImage

    (2011)
  • Z. Lao et al.

    Computer-assisted segmentation of white matter lesions in 3D MR images using support vector machines

    Acad. Radiol.

    (2008)
  • M. Martin-Fernandez et al.

    Sequential anisotropic multichannel Wiener filtering with Rician bias correction applied to 3D regularization of DWI data

    Med. Image Anal.

    (2009)
  • K. Nakamura et al.

    Segmentation of brain magnetic resonance images for measurement of gray matter atrophy in multiple sclerosis patients

    NeuroImage

    (2009)
  • C. Pollo et al.

    Segmentation of brain structures in presence of a space-occupying lesion

    NeuroImage

    (2005)
  • M. Prastawa et al.

    A brain tumor segmentation framework based on outlier detection

    Med. Image Anal.

    (2004)
  • K. Rehm et al.

    Putting our heads together: a consensus approach to brain/non-brain segmentation in T1-weighted MR volumes

    NeuroImage

    (2004)
  • D. Rey et al.

    Automatic detection and segmentation of evolving processes in 3D medical images: application to multiple sclerosis

    Med. Image Anal.

    (2002)
  • A. Rovira et al.

    MR in the diagnosis and monitoring of multiple sclerosis: an overview

    Eur. J. Radiol.

    (2008)
  • S. Saha et al.

    A new point symmetry based fuzzy genetic clustering technique for automatic evolution of clusters

    Inf. Sci.

    (2009)
  • M.L. Seghier et al.

    Lesion identification using unified segmentation-normalisation models and fuzzy clustering

    NeuroImage

    (2008)
  • D.D. Sha et al.

    Towards automated enhancement, segmentation and classification of digital brain images using networks of networks

    Inf. Sci.

    (2001)
  • M. Shah et al.

    Evaluating intensity normalization on MRIs of human brain with multiple sclerosis

    Med. Image Anal.

    (2011)
  • D.W. Shattuck et al.

    Magnetic resonance image tissue classification using a partial volume model

    NeuroImage

    (2001)
  • S. Shen et al.

    An improved lesion detection approach based on similarity measurement between fuzzy intensity segmentation and spatial probability maps

    Magn. Reson. Imag.

    (2010)
  • N. Shiee et al.

    A topology-preserving approach to the segmentation of brain images with multiple sclerosis lesions

    NeuroImage

    (2010)
  • J. Souplet et al.

    Revue des approches de segmentation des lesions de sclerose en plaques dans les sequences conventionnelles IRM

    Rev. Neurol.

    (2009)
  • F. Admiraal-Behloul et al.

    Fully automatic segmentation of white matter hyperintensities in MR images of the elderly

    NeuroImage

    (2005)
  • L. Aït-Ali, S. Prima, P. Hellier, B. Carsin, G. Edan, C. Barillot, STREM: a robust multidimensional parametric method...
  • A. Akselrod-Ballin et al.

    Automatic segmentation and classification of multiple sclerosis in multichannel MRI

    IEEE Trans. Biomed. Eng.

    (2009)
  • B. Alfano et al.

    Automated segmentation and measurement of global white matter lesion volume in patients with multiple sclerosis

    J. Magn. Reson. Imag.

    (2000)
  • P. Anbeek, K. Vincken, M. Viergever, Automated MS-lesion segmentation by k-nearest neighbor classification, in: Grand...
  • Asclepios-SepINRIA. <http://www-sop.inria.fr/asclepios/software/SepINRIA/>, 2008 (accessed...
  • M. Bach-Cuadra et al.

    Comparison and validation of tissue modelization and statistical classification methods in T1-weighted MR brain images

    IEEE Trans. Med. Imag.

    (2005)
  • M. Bach-Cuadra et al.

    Atlas-based segmentation of pathological MR brain images using a model of lesion growth

    IEEE Trans. Med. Imag.

    (2004)
  • F. Bagnato et al.

    Evolution of T1 black holes in patients with multiple sclerosis imaged monthly for 4 yr

    Brain

    (2003)
  • B.J. Bedell et al.

    Automatic segmentation of gadolinium-enhanced multiple sclerosis lesions

    Magn. Reson. Med.

    (1998)
  • J.C. Bezdek

    Pattern Recognition With Fuzzy Objective Function Algorithms

    (1981)
  • J.C. Bezdek et al.

    Review of MR image segmentation techniques using pattern recognition

    Med. Phys.

    (1993)
  • S. Bricq, C. Collet, J. Armspach, Ms lesion segmentation based on hidden Markov chains, in: Grand Challenge Work.:...
  • D.L. Collins et al.

    Design and construction of a realistic digital brain phantom

    IEEE Trans. Med. Imag.

    (1998)
  • D. Comaniciu et al.

    Mean shift: a robust approach toward feature space analysis

    IEEE Trans. Pattern Anal. Machine Intel.

    (2002)
  • Cited by (0)

    View full text