Original contributionAutomatic segmentation of cerebral white matter hyperintensities using only 3D FLAIR images
Introduction
White matter hyperintensities (WMHs) are diffuse white matter abnormalities that appear with high intensities in T2-weighted magnetic resonance (MR) images. Although the pathogenesis of WMHs is not yet completely understood, these lesions are often associated with chronic cerebral ischemia, in particular with microvascular lesions originated by small vessel atherosclerosis [1]. They occur often in the elderly [2], [3], [4], [5] and have been shown to predict an increased risk of stroke, cognitive decline and death [6].
The analysis of the real influence of WMHs on the development of dementia requires clinical studies involving large patient cohorts. Also, an accurate description of the location, shape and volume of the WMHs is necessary. Typically, WMHs are classified according to visual scales, such as the Scheltens scale or the Fazekas scale [7]. However, the results obtained by these visual scales are seldom comparable [8]. In addition, they have been shown to be little sensitive to clinical group differences [9]. Finally, they offer only a qualitative description of the WMHs, originating high intra- and inter-subject variabilities [10].
A quantitative and more reliable way of assessing WMHs is by manually determining the lesion volumes. However, for three-dimensional data this typically requires a slice-by-slice analysis, making the whole process cumbersome and time-consuming for the neuroradiologist. Also, the intra- and inter-rater variability have been reported to be high [11]. Clinical studies with hundreds of patients require, therefore, automated and robust segmentation methods.
Several methods have been proposed to automatically segment WMHs from MRI images, most of them using various types of MRI modalities [12], [13], [14]. The use of multimodal data presents several disadvantages. Namely, the acquired datasets must be coregistered, making the segmentations computationally intensive and more prone to errors. In particular, motion artifacts are seen frequently in the MRI data from elderly patients, who are often not able to lie still during the whole acquisition period. This represents a serious limitation for the registration algorithms and can negatively influence the outcomes [15], [16].
Other methods have been specifically designed to segment multiple sclerosis (MS) lesions [17], [18]. Although MS lesions look similar to vascular-related WMHs in MR images, the spatial distribution of the lesions is often very different, with MS lesions occurring commonly in the corpus callosum and being symmetrically distributed in the brain, unlike the vascular WMHs [19].
WMHs are characterized by a larger T2 relaxation rate due to increased tissue water content and degradation of myelin [15]. Fast fluid-attenuated inversion-recovery (FLAIR) is a T2-weighted MR modality in which the cerebrospinal fluid (CSF) signal is attenuated. In FLAIR images, WMHs are characterized by an intensity range that only partially overlaps with that of normal brain regions, making this MRI modality well suited for lesion segmentation purposes [20].
Despite being the preferred imaging modality used by neuroradiologists to assess WMHs in the clinical setting, FLAIR has seldom been used alone in the automatic detection of these lesions [15], [16].
In [15], the authors determined an optimal FLAIR intensity threshold to separate WMHs from normal brain tissue, based on the analysis of the image histograms on a training set. More recently, Ong et al. [21] have applied an outlier detection approach to find this optimal threshold, followed by a false positive correction step that uses the co-registered T1-weighted image. Similarly, de Boer et al. [14] determined the optimal intensity threshold on a training set and used the T1-weighted image to ensure the detected lesions were all within the white matter.
Applying a threshold allows only for crisp segmentation and does not account for the Partial Volume Effect (PVE) that is present in MR images. Having that in mind, Khademi et al. have proposed a segmentation method that allows for fuzzy segmentation and is based on a PVE model in FLAIR images [16].
In the methods described above, only the voxel intensity information is considered. However, it has been recognized that this makes methods highly sensitive to noise. In particular, boundary detection becomes problematic in noisy images. Furthermore, the common assumption that the voxel intensities are independent does not hold in practice. In reality, and intuitively, we can expect a certain voxel's value to be affected by those in its neighborhood [22], [23].
In this work, we propose a WMH segmentation method that uses solely FLAIR images. It is based on a modified Gaussian mixture model (GMM) that incorporates neighborhood information, followed by a false positive correction step, where common FLAIR artifacts [24] are eliminated from the segmentation.
Gaussian mixture models (GMM), estimated by the expectation-maximization (EM) algorithm, have been widely used in brain image segmentation [25], [26]. They provide a statistical description of the voxels' intensities and allow for fuzzy classification [27]. Because the traditional GMM-EM method is based only on intensity information, we use a modified GMM-EM method, initially proposed in [23], that considers additional contextual information. All initialization parameters are derived from the FLAIR image histogram.
We compare the performance of the proposed method with other unimodal approaches. For each method, the optimal parameters are determined using a training set that is retrieved randomly from our patient database. Evaluation is performed using the remaining patient datasets against the manual segmentation performed by an experienced neuroradiologist. Finally, we apply the method to a publicly available dataset of MS patients and compare the obtained performance results with those by multimodal segmentation methods and with the human expert.
Section snippets
Methods
Fig. 1 shows the general overview of our method.
The raw FLAIR image is first preprocessed to remove the skull and to correct for bias field inhomogeneities. Subsequently, a context-sensitive GMM is applied to the brain image and the resulting WMH probability class is thresholded. Finally, the existing FLAIR artifacts (located at the interface between the cerebrospinal fluid and the gray matter and inside the ventricles – red pixels in the last figure) are eliminated by morphological processing
Data
Forty datasets were retrieved from a large database of a cognition study with MCI and control subjects carried out at the University Hospital of Essen, Germany. From these 40 subjects, 15 correspond to stable normal controls, 14 to stable amnestic-MCI subjects, 8 to MCI subjects who have progressed to dementia and 3 to normal subjects who have declined to amnestic-MCI. The age of the subjects is 74.7 ± 4.3 (range 62–82).
Three-dimensional isotropic FLAIR images are utilized in this study (1.5 T
Conclusion
In this work, we present a method to automatically segment WMHs using only 3D FLAIR images. It uses a context-sensitive Gaussian mixture model to obtain class probabilities, followed by crisp segmentation and artifact correction. Unlike the majority of the existing approaches (to the best of our knowledge), our method requires no additional MRI modalities nor atlases, thereby shortening the acquisition time, avoiding the need for co-registrations and allowing for near real-time analysis.
References (39)
- et al.
Quantification of MRI lesion load in multiple sclerosis: a comparison of three computer-assisted techniques
Magn Reson Imaging
(1996) - et al.
Automatic segmentation of different-sized white matter lesions by voxel probability estimation
Med Image Anal
(2004) - et al.
Fully automatic segmentation of white matter hyperintensities in MR images of the elderly
Neuroimage
(2005) - et al.
White matter lesion extension to automatic brain tissue segmentation on MRI
Neuroimage
(2009) - et al.
A topology-preserving approach to the segmentation of brain images with multiple sclerosis lesions
Neuroimage
(2010) - et al.
Spatial decision forests for MS lesion segmentation in multi-channel magnetic resonance images
Neuroimage
(2011) - et al.
Automatic white matter lesion segmentation using an adaptive outlier detection method
Magn Reson Imaging
(2012) - et al.
A vectorial image soft segmentation method based on neighborhood weighted Gaussian mixture model
Comput Med Imaging Graph
(2009) - et al.
Segmentation of age-related white matter changes in a clinical multi-center study
Neuroimage
(2008) - et al.
The significance of cerebral white matter abnormalities 100 years after Binswanger's report. A review
Stroke
(1995)
Prevalence of cerebral white matter lesions in elderly people: a population based magnetic resonance imaging study. the Rotterdam Scan Study
J Neurol Neurosurg Psychiatry
Regional variability in the prevalence of cerebral white matter lesions: an MRI study in 9 European countries (CASCADE)
Neuroepidemiology
White matter hyperintensities on MRI in the neurologically nondiseased elderly. Analysis of cohorts of consecutive subjects aged 55 to 85 years living at home
Stroke
Prevalence of white matter hyperintensities in a young healthy population
J Neuroimaging
The clinical importance of white matter hyperintensities on brain magnetic resonance imaging: systematic review and meta-analysis
BMJ
White matter changes on CT and MRI: an overview of visual rating scales. European task force on age-related white matter changes
Eur Neurol
Variable agreement between visual rating scales for white matter hyperintensities on MRI: comparison of 13 rating scales in a poststroke cohort
Stroke
Impact of white matter hyperintensities scoring method on correlations with clinical data: the LADIS study
Stroke
Measuring longitudinal white matter changes: comparison of a visual rating scale with a volumetric measurement
AJNR Am J Neuroradiol
Cited by (50)
Automatic detection of white matter hyperintensities via mask region-based convolutional neural networks using magnetic resonance images
2022, Deep Learning for Medical Applications with Unique DataSegmentation of white matter lesions in multicentre FLAIR MRI
2021, Neuroimage: ReportsFLAIR-only joint volumetric analysis of brain lesions and atrophy in clinically isolated syndrome (CIS) suggestive of multiple sclerosis
2021, NeuroImage: ClinicalCitation Excerpt :Volumetric techniques for total lesion load and brain volume quantification have been developed in the research and clinical trial settings, where image acquisition is more homogeneous and multiple contrasts are available (Lindig et al., 2018; Danelakis et al., 2018). In general, lesion segmentation techniques rely on the availability of multi-contrast source image data sets, i.e. requiring both T1- and T2-weighted (e.g. T2-FLAIR) images, with automated techniques typically reliant on isotropic three dimensional (3D) acquisitions but manual delineations often performed on two dimensional (2D) acquisitions (Simões et al., 2013; de Boer et al., 2009). Brain volume quantification solutions typically require a 3D T1-weighted image dataset.
A survey on U-shaped networks in medical image segmentations
2020, NeurocomputingCitation Excerpt :This approach has a drawback is that although 2D slice makes full use of the information in the entire 2D slice, it ignores the relationship between several adjacent slices and loses the correlation between global slice sequences. ( 2) Using 3D images as inputs [26,73,116–120]. Compared with using 2D images, using 3D images as input can obtain more semantic information, which helps improve the performance of segmentation models.
An improved algorithm of white matter hyperintensity detection in elderly adults
2020, NeuroImage: ClinicalAre multi-contrast magnetic resonance images necessary for segmenting multiple sclerosis brains? A large cohort study based on deep learning
2020, Magnetic Resonance ImagingCitation Excerpt :FPR appears to be a more sensitive measure for evaluating smaller lesion segmentation as seen in its improvement when PD, T1 and T2 images were included with FLAIR. While a few studies investigated the quality of WM hyperintense lesion segmentations using FLAIR alone [29–31], few compared the segmentation performance using different input sequences in MS. Using 3D CNN model, Brosch et al. [32] investigated the effect of different sequences on MS lesion segmentation. Specifically, these authors evaluated three contrast combinations: T1 + FLAIR, T1 + T2 + PD, and T1 + T2 + FLAIR + PD on lesion segmentation.