Elsevier

NeuroImage

Volume 104, 1 January 2015, Pages 287-300
NeuroImage

2D harmonic filtering of MR phase images in multicenter clinical setting: Toward a magnetic signature of cerebral microbleeds

https://doi.org/10.1016/j.neuroimage.2014.08.024Get rights and content

Highlights

  • We introduce a fast 2D processing technique of MR phase images (~ 2.2 s).

  • Method includes unwrapping and harmonic filtering to yield internal field variations.

  • Comparison showed higher performance among state-of-the art filtering approaches.

  • Resulting internal field maps aim to better discriminate cerebral microbleeds.

  • We demonstrate its applicability and robustness on multicenter clinical 2D datasets.

Abstract

Cerebral microbleeds (CMBs) have emerged as a new imaging marker of small vessel disease. Composed of hemosiderin, CMBs are paramagnetic and can be detected with MRI sequences sensitive to magnetic susceptibility (typically, gradient recalled echo T2* weighted images). Nevertheless, their identification remains challenging on T2* magnitude images because of confounding structures and lesions. In this context, T2* phase image may play a key role in better characterizing CMBs because of its direct relationship with local magnetic field variations due to magnetic susceptibility difference. To address this issue, susceptibility-based imaging techniques were proposed, such as Susceptibility Weighted Imaging (SWI) and Quantitative Susceptibility Mapping (QSM). But these techniques have not yet been validated for 2D clinical data in multicenter settings. Here, we introduce 2DHF, a fast 2D phase processing technique embedding both unwrapping and harmonic filtering designed for data acquired in 2D, even with slice-to-slice inconsistencies. This method results in internal field maps which reveal local field details due to magnetic inhomogeneity within the region of interest only. This technique is based on the physical properties of the induced magnetic field and should yield consistent results. A synthetic phantom was created for numerical simulations. It simulates paramagnetic and diamagnetic lesions within a ‘brain-like’ tissue, within a background. The method was evaluated on both this synthetic phantom and multicenter 2D datasets acquired in standardized clinical setting, and compared with two state-of-the-art methods. It proved to yield consistent results on synthetic images and to be applicable and robust on patient data. As a proof-of-concept, we finally illustrate that it is possible to find a magnetic signature of CMBs and CMCs on internal field maps generated with 2DHF on 2D clinical datasets that give consistent results with CT-scans in a subsample of 10 subjects acquired with both modalities.

Introduction

Cerebral microbleeds (CMBs) were initially detected in histopathological studies of patients with small vessel disease (Cordonnier and van der Flier, 2011, Cordonnier, 2011, Greenberg et al., 2009). They were described as small foci of chronic blood products in normal brain tissue (Greenberg et al., 2009, Poels et al., 2010, Van der Flier and Cordonnier, 2012). Their size may vary from very small (~ 2 mm) to large lesions (~ 10 mm), while larger lesions are assumed to be more spread hemorrhages. CMB characterization is of interest within the study of vascular dementia and Alzheimer's disease (AD) (Cordonnier and van der Flier, 2011). However, CMB identification using MRI remains challenging (Cordonnier and van der Flier, 2011, Greenberg et al., 2009).

CMBs are made of hemosiderin which is a strongly superparamagnetic iron–storage complex (Cordonnier, 2011), whereas brain parenchyma is diamagnetic. Thus, their strong susceptibility difference with brain parenchyma makes CMBs foci appear as magnetic inclusions, causing local magnetic field inhomogeneity such as would be created by a unit dipole. At the voxel level, this field inhomogeneity leads to intravoxel phase dispersion and strong T2*-contrast. Its detection is thus commonly based on Gradient Recalled Echo (GRE) T2*-weighted magnitude images, in which CMBs are visible as a loss of signal (hypointensity). However, their appearance on these sequences is sensitive to imaging parameters such as echo time (TE) and B0 field strength; clinical interpretation is thus made difficult by the resulting blooming artifacts. Furthermore, blood vessels and cerebral micro-calcifications (CMCs) also have strong T2* effects and can be misidentified as CMBs. While localization may help identification, discrimination commonly requires additional T1-weighted or T2-weighted images, or even a CT scan for CMCs (Yamada et al., 1996).

In order to overcome some limitations in CMB identification, the phase image could also be considered. Usually discarded, the phase is available at no extra acquisition time. Being proportional to the local resonance frequency, phase directly reflects magnetic field inhomogeneity. Using phase information could allow increasing both specificity and sensitivity in CMB detection. For example, calcifications are more diamagnetic than brain parenchyma and the induced magnetic field perturbation is opposed to that of paramagnetic CMBs; this difference should be accessible through phase information (Gronemeyer et al., 1992, Gupta et al., 2001, Schweser et al., 2010, Yamada et al., 1996). As for sensitivity, phase contrast strongly derives from susceptibility distribution and enhanced contrast could be expected on phase images between paramagnetic CMBs and parenchyma.

While phase is sensitive to local susceptibility variations, its analysis is not straightforward because of phase wrapping and strong background effects, as well as the complex magnetic field-to-source relationship. Indeed, reconstructed phase appears “wrapped”, as it is only defined within [− π, π]; it thus requires the use of unwrapping techniques to recover a continuous phase information (Feng et al., 2013). Additionally, local variations of interest may be orders of magnitude lower than those related to the background field inhomogeneity, which is dominated by the air–tissue interface, thus requiring efficient filtering algorithms to extract the contribution of the internal local field inhomogeneity pattern (De Rochefort et al., 2010b).

The first studies exploring the use of GRE phase images to discriminate between calcified and iron-laden tissues relied only on raw phase images (Yamada et al., 1996). To further enhance detection sensitivity for small inclusions, background contributions have been suppressed using different high pass filters (Gronemeyer et al., 1992, Gupta et al., 2001, Wu et al., 2009, Yamada et al., 1996). Combining phase and magnitude images, such as in susceptibility-weighted imaging (SWI), has already allowed enhancing detection sensitivity for paramagnetic structures such as veins or hemorrhages (Goos et al., 2011, Haacke and Reichenbach, 2011, chap. 1; Haacke et al., 2009, Nandigam et al., 2009, Reichenbach et al., 1997). Recent advances in the understanding of magnetic field distortions yielded more adapted phase processing techniques. Indeed, quantitative susceptibility mapping (QSM) is based on the reconstruction of magnetic susceptibility maps from an observed magnetic field perturbation (De Rochefort et al., 2010b, Liu et al., 2012, Schweser et al., 2010). These approaches have allowed to push further the limits of background field removal and solve the “field-to-source” inverse problem (Langkammer et al., 2012, Li et al., 2011, Schweser et al., 2011, Schweser et al., 2012b), enabling to differentiate calcifications from hemorrhages (De Rochefort et al., 2010b, Deistung et al., 2006, Reichenbach and Haacke, 2001, Schweser et al., 2010) and to provide improved CMB detection sensitivity and contrast as compared to GRE magnitude images (Klohs et al., 2011, Liu et al., 2012). These latter approaches generally rely on an inverse filter design based on complex post-processing methods; computing strategies currently remain under investigation. Furthermore, both SWI and QSM were designed for being applied to full 3D dataset and phase unwrapping and background field removal are necessary preprocessing steps for both methods.

To recover the internal field, several background field filtering techniques have been proposed. Assuming that background field variation mostly contains low frequency components within the region of interest while that of internal field contains high frequency components, low pass filtering using Gaussian (Hammond et al., 2008) or box kernel (Rauscher et al., 2003) or low order polynomial fitting (Deistung et al., 2008, Duyn et al., 2007) were first proposed. More recently, approaches relying on fitting external sources to internal field were proposed, using either highly-constrained model-based distributions (De Rochefort et al., 2008, De Rochefort et al., 2010b, Neelavalli et al., 2009, Wharton et al., 2010), or fitting with more degrees of freedom such as in Projection onto Dipole Field (PDF) (Liu et al., 2011). The PDF approach has demonstrated efficient estimation of background field in an internal region of interest (ROI), but displayed remaining border artefacts (Liu et al., 2011). Finally, harmonic filtering techniques, such as Sophisticated Harmonic Artifact Reduction for Phase (SHARP) (Schweser et al., 2011), rely on the harmonic property of the background field inside a ROI, leading to a new class of Laplace based filters (Schweser et al., 2011, Schweser et al., 2012a).

In the context of multicenter clinical studies, data from various manufacturers and models have to be analyzed jointly, even though phase image properties may differ between acquisition sites. Subtle differences in pulse sequence characteristics, coil sensitivity profiles, localization methods, phase reconstruction algorithms and other site/manufacturer specific characteristics may combine to produce significant variation in final measurements. These differences must be taken into account to improve final pooled analyses. Furthermore, standard multi-slice 2D scan may result in inconsistent slice-to-slice field maps. These linear terms were observed experimentally on clinical datasets (Lee et al., 2013, Tam et al., 2009). They may result from different 2D-based processes (shimming, motion, breathing-related artifact, normalization…), either at the acquisition or reconstruction levels, and lead to inconsistent phase maps between slices.

Here, we propose a filter design acting directly on the default reconstructed phase images to estimate internal field maps. This filter relies on a fast and robust 2D harmonic filtering (2DHF) approach that includes unwrapping, background field removal and additional linear artifact (due to slice-to-slice inconsistencies) correction at the same time. The method aims at being applicable on 2D datasets acquired in clinical settings in a multicenter framework.

Phase was long considered as unreliable information due to, first, its “non-local” nature and, second, its dependency to the two pre-processing steps described above (Schweser et al., 2010). The first issue is related to the non-local relationship between magnetic susceptibility distribution and phase. QSM may overcome this issue through the source reconstruction step but some approaches are computationally expensive and not straightforward to apply in multicenter settings. However, for clinical purpose, the main focus is on the type of lesion, namely diamagnetic or paramagnetic; “non-local field perturbation” may thus not be a limitation for clinical application based on internal field maps only. For the second issue, recent techniques such as SHARP were shown to allow robust pre-processing of phase images. 2DHF can be considered as a 2D version of SHARP, introduced as a 3D filtering technique in Schweser et al. (2011).

The remainder of this article is organized as follows. The multicenter dataset used for validation is first presented, followed by a detailed description of the filtering method. Numerical simulation used for synthetic evaluation and two state-of-the-art filtering approaches used in a comparison study are then presented, as well as implementation issues. Evaluation and comparison results on numerical simulation and patient data are then shown, followed by a proof of concept illustration for the ability of 2DHF to define a magnetic signature for CMBs and CMCs on multicenter 2D datasets acquired in patients with memory impairment.

Section snippets

Materials and methods

Data on which the method was evaluated will be presented first together with acquisition details. The filtering method will then be described as well as the state-of-the-art filtering approaches and simulation used for validation.

Results

Simulation results for the 2DHF method will first be shown, together with a comparison with the two state-of-the-art methods HPF and PDF. Results and comparison between methods on clinical data will then be detailed, followed by a proof-of-concept for the discrimination between CMBs and CMCs based on internal field maps obtained with 2DHF.

Discussion

This work presents a new efficient tool for background field removal in clinical multicenter setting. Unwrapping and local field estimation were simultaneously performed using a 2D version of a harmonic filter (2DHF), applied in Fourier domain. The 2D harmonic filter removes background effects while preserving local phase variations. This method showed good performance in retrieving fine 3D coherent details on 2D datasets on simulated and clinical images and allowed to identify a magnetic

Acknowledgments

The research leading to these results has received funding from the program “Investissements d'avenirANR-10-IAIHU-06.

The Memento study is undertaken through the sponsorship of “Bordeaux CHU” and the financial support of “Fondation Plan Alzheimer”. The funding sources had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

The

References (57)

  • F. Schweser et al.

    Quantitative imaging of intrinsic magnetic tissue properties using MRI signal phase: an approach to in vivo brain iron metabolism?

    NeuroImage

    (2011)
  • F. Schweser et al.

    Quantitative susceptibility mapping for investigating subtle susceptibility variations in the human brain

    NeuroImage

    (2012)
  • M. Shive et al.

    Quantitative 3D MRI as a valid endpoint for randomized clinical trials in cartilage repair and its correlation with repair tissue collagen architecture

    Arthroscopy

    (2013)
  • R.C. Tam et al.

    Detection and measurement of coverage loss in interleaved multi-acquisition brain MRIs due to motion-induced inter-slice misalignment

    Med. Image Anal.

    (2009)
  • W.M. Van der Flier et al.

    Microbleeds in vascular dementia: clinical aspects

    Exp. Gerontol.

    (2012)
  • M. Ayaz et al.

    Imaging cerebral microbleeds using susceptibility weighted imaging: one step toward detecting vascular dementia

    J. Magn. Reson. Imaging

    (2010)
  • T.E. Conturo et al.

    Signal-to-noise in phase angle reconstruction: dynamic range extension using phase reference offsets

    Magn. Reson. Med.

    (1990)
  • C. Cordonnier

    Brain microbleeds: more evidence, but still a clinical dilemma

    Curr. Opin. Neurol.

    (2011)
  • C. Cordonnier et al.

    Brain microbleeds and Alzheimer’s disease: innocent observation or key player?

    Brain

    (2011)
  • L. De Rochefort et al.

    Quantitative MR susceptibility mapping using piece-wise constant regularized inversion of the magnetic field

    (2008)
  • L. De Rochefort et al.

    Quantitative susceptibility map reconstruction from MR phase data using bayesian regularization: validation and application to brain imaging

    Magn. Reson. Med.

    (2010)
  • L. De Rochefort et al.

    Quantitative susceptibility mapping in vivo in the rat brain

  • A. Deistung et al.

    Susceptibility weighted imaging at ultra high magnetic field strengths: theoretical considerations and experimental results

    Magn. Reson. Med.

    (2008)
  • J.H. Duyn et al.

    High-field MRI of brain cortical substructure based on signal phase

    Proc. Natl. Acad. Sci.

    (2007)
  • W. Feng et al.

    Catalytic multiecho phase unwrapping scheme (CAMPUS) in multiecho gradient echo imaging: removing phase wraps on a voxel-by-voxel basis

    Magn. Reson. Med. Off. J. Soc. Magn. Reson. Med. Soc. Magn. Reson. Med.

    (2013)
  • J.D.C. Goos et al.

    Clinical relevance of improved microbleed detection by susceptibility-weighted magnetic resonance imaging

    Stroke J. Cereb. Circ.

    (2011)
  • S.M. Gregoire et al.

    The Microbleed Anatomical Rating Scale (MARS): reliability of a tool to map brain microbleeds

    (2009)
  • S.A. Gronemeyer et al.

    MR imaging detection of calcified intracranial lesions and differentiation from iron-laden lesions

    J. Magn. Reson Imaging

    (1992)
  • Cited by (15)

    • Clinical Integration of Quantitative Susceptibility Mapping Magnetic Resonance Imaging into Neurosurgical Practice

      2019, World Neurosurgery
      Citation Excerpt :

      GRE phase-based imaging has been explored to overcome SWI's inability to differentiate diamagnetic from paramagnetic sources.32 Through phase-filtering techniques,33 it is now possible to rapidly differentiate the sign of susceptibility sources on internal field maps: positive for paramagnetic iron and negative for diamagnetic calcification. Although these advanced filtering approaches are superior compared with SWI and T2*w imaging,34 they are still not able to provide precise susceptibility source localization.

    • Improved cerebral microbleeds detection using their magnetic signature on T2*-phase-contrast: A comparison study in a clinical setting

      2017, NeuroImage: Clinical
      Citation Excerpt :

      In fact, these inclusions behave like small magnetic dipoles and thus create magnetic field patterns similar to those of dipoles; this dipolar magnetic field appears as a ring-like effect in the axial plane in IFM. The sign pattern depends on the lesion susceptibility class (paramagnetic or diamagnetic) and allows to discriminate between CMBs and cerebral micro-calcifications related mimics (Kaaouana et al., 2015). Overall, IFM thus improve sensitivity (with respect to magnitude image) and decrease FPR (with respect to SWI-mIP).

    • Fast quantitative susceptibility mapping using 3D EPI and total generalized variation

      2015, NeuroImage
      Citation Excerpt :

      The clinical value of QSM has not yet been fully explored but holds great promise for vascular, inflammatory and neurodegenerative diseases of the brain (Wang and Liu, 2014). To date, QSM has been shown to differentiate micro-bleeds from calcifications which otherwise appear similar on T2*-weighted images (Kaaouana et al., 2014; Liu et al., 2011b; Schweser et al., 2010a). QSM allows better delineation of deep brain structures from adjacent tissue than T2*-weighted images (Deistung et al., 2013; Eskreis-winkler et al., 2013; Schäfer et al., 2012); can be used to assess myelin content and the extent of loss during demyelination (Argyridis et al., 2013; Stüber et al., 2014; Wharton and Bowtell, 2014); and has the potential to map oxygen saturation along venous vasculature (Fan et al., 2013).

    View all citing articles on Scopus
    View full text