HAL Id: hal-01088726
https://hal.archives-ouvertes.fr/hal-01088726
Submitted on 13 Jan 2015
HAL is a multi-disciplinary open access archive for the deposit and dissemination of sci- entific research documents, whether they are pub- lished or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers.
L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.
Validation of 3D spino-pelvic muscle reconstructions based on dedicated MRI sequences for fat-water
quantification
Bertrand Moal, J-G Raya, Erwan Jolivet, F. Schwab, B. Blondel, Virginie Lafage, Wafa Skalli
To cite this version:
Bertrand Moal, J-G Raya, Erwan Jolivet, F. Schwab, B. Blondel, et al.. Validation of 3D spino-pelvic muscle reconstructions based on dedicated MRI sequences for fat-water quantifica- tion. Innovation and Research in BioMedical engineering, Elsevier Masson, 2014, 35, pp.119-127.
�10.1016/j.irbm.2013.12.011�. �hal-01088726�
Sc ience Arts & Mét iers (SAM)
is an open access repositorythat collectsthe work of Arts et Métiers ParisTech researchers and makesitfreely available overthe web where possible.
Thisis an author-deposited version publishedin: http://sam.ensam.eu HandleID:.http://hdl.handle.net/10985/9014
To cite this version :
Bertrand MOAL, Jose G. RAYA, Erwan JOLIVET, Franck SCHWAB, Benjamin BLONDEL, Virginie LAFAGE, Wafa SKALLI - Validation of 3D spino-pelvic muscle reconstructions based on dedicated MRI sequencesforfat-water quantification -IRBM - Vol. 35, p.119-127 - 2014
Any correspondence concerningthis service should be senttothe repository Administrator: archiveouverte@ensam.eu
Va l ida t ion of 3D sp ino-pe lv ic musc le recons truc t ions based on ded ica ted MRI sequences for fa t-wa ter quan t ifica t ion
B . Moa l
a,c, J .G . Raya
b, E . Jo l ive t
c,F .Schwab
a, B . B londe l
a,V .Lafage
a,∗,W . Ska l l i
caSpine Division, NYU Hospitalfor Joint Diseases, New York University Langone Medical Center, 306 E. 15th St., Suite 1F, New York, NY, 10003 USA
bCenterfor BiomedicalImaging, Department of Radiology, New York University Langone Medical Center, 550,FirstAvenue, New York, NY, 10016 USA
cLaboratoire de biomécanique, arts et métiers ParisTech, 151, boulevard del’Hôpital, 75013 Paris, France
Abstract
Objectives. – To evaluate a protocol, includingMRI acquisition with dedicated sequences for fat-water quantification and semi-automatic segmen- tation,for 3D geometry measurement and fat infiltrationof key muscles of thespino-pelvic complex.
Materials and methods. – MRI protocol: twoaxial acquisitions from thethoraco-lumbarregion tothepatella were obtained: one T1-weighted and one based on theDixon method, permitted toevaluate theproportion of fat insideeach muscle. Muscle reconstruction: with Muscl’X software, 3D reconstructions of 18 muscles or groups of muscles were obtained identifyingtheircontours on a limitednumber of axial images3D references were obtained only on T1 acquisitions identifyingthecontour of themuscles on all axial images.Evaluation: for twovolunteers, threeoperators completed reconstructions threetimesacross threesessions. Each reconstruction was projected on thereference tocalculate the‘point tosurface’
error.Mean and maximal axial section, muscle volume, and muscle lengthcalculated from thereconstructions were compared toreference values, and intra-and inter-operatorvariability for thoseparameters were evaluated.
Results. – 2xRMS ‘point tosurface’ error was below 3 mm, on average. The agreement between thetwomethods was variable between muscles [–4.50; 8.00%] for themean axial section, thelengthand thevolume. Intra- and inter-operatorvariability were lessthan5% and comparison of variability for theFat and T1 reconstructions did not reveal any significant differences.
Discussion. – Excellent inter-and intra-operatorreliability was demonstrated for 3D muscular reconstruction using theDPSO method and Dixon imagesthatallowed generation of patient-specific musculoskeletal models.
1. Introduction
The muscular system plays an essential role in the main- tenance of postural balance; however, in clinical practice as well asresearch,investigationintotherelationship betweenthe muscular system and postural pathologies such as adult spinal deformity, hasbeenlimited.
Thecurrentlack of knowledgerelatedtosofttissuestabilizers may be attributed to the absence of a relevant tool to evaluate the muscular system as a whole. The large number of muscles involvedin postural maintenance makes a global analysis diffi- cult and time consuming. In the case of adult spinal deformity (ASD),recentresearch has highlightedthecriticalrole ofsagittal spino-pelvic alignment in patient-reported pain and disability
∗Corresponding author.
E-mail address:virginie.lafage@gmail.com(V. Lafage).
[1–5]. Therefore,keymusclesinvolvedin pelvic-positioningand lumbar-stabilization areattheforefront ofresearch needs.
Preliminaryefforts have been directedtowards understanding the muscular envelope of the spine using histological analyses [6–9], measurement of muscularstrength[10–12],and measure- ment of electromyographical signals [13,14]. However, these approaches are not adaptedfor study of alarge number of mus- cle groups. Otherstudies, usingimagingsuchas MRI or CTscan, correlated measurement ofthe muscle cross-sectional areas(via ultrasound, CT-scan or MRI)[15–23], or measurement of mus- cular density(via CT-scan or MRI)[10–12,17–21,23–25], with chronic back pain orspinalsurgery outcome. Thelimitations of these approaches lie in the difficulty to represent variability in volumeofan entire muscle[26].
Toovercome these limitations, Jolivet et al.[27,28]devel- oped a method ofthree-dimensional muscle reconstruction via segmentation ofa small number of axial images (MRI or CT- scan). This method, based on the deformation of a parametric
Fig. 1. Examples of a T1(left), water(center), andFat(right)image.
specific object(DPSO), has beensuccessfullyimplemented with CT scans for analysis of musclesinvolvedin knee motion[29]
and muscle groupsaroundthe hipjoint[28]. The CT-scan modal- ity presents two advantages: good contrast quality for muscle segmentationand good reliability in terms offatand muscle density.CTscansallowfor bothareproducibleanalysis of mus- cle geometry and a quantified evaluation offatinfiltration[28].
Nevertheless, the radiation exposure from CT scans renders it unacceptable as a tool for studies involving ASD patients who are already frequently subjected to radiographic examination. Notably,the DPSO method has also been performed using MRI T1 sequences[29]. However, the inhomogeneity in the mag- netic fieldapplied did notallowanaccurate quantification ofthe fatinfiltration(withoutfatinfiltration,the muscle volumeis an incomplete descriptor).
In order to avoid the problem of inhomogeneity and obtain an accurate quantification offatinfiltration, Dixon et al.[30]
developedaspecificacquisitionsequence wheretwoimagesare obtained: onein whichtheintensity of eachvoxelis correlated with the quantity offatand the other in which the intensity of eachvoxelis correlated withthe quantity of water. This method wasthenimproved byGloveret al.[31].Toour knowledge, the feasibility of 3D muscular reconstruction on MRI with the Dixon methodwasnot studied.
The objective ofthisstudywastoevaluatethefeasibility ofan MRI protocol with dedicatedsequencesforfat-water quantifica- tiontoassessthe 3D geometryand homogeneity(fatinfiltration) ofkeymusclesinthe spino-pelvic complex.
2. Methods 2.1. Subject sample
Twoasymptomaticfemale adult volunteers wereincludedin this pilot study: volunteer A (35 years, 68 kg) and volunteer B (38 years, 91 kg).
2.2. MRI acquisition
MRI was performed on a 3 T whole-body scanner (Magnetom Verio, Siemens Healthcare, Erlangen, Germany) using a 24-channel spine matrix coil and three 4-channel flex coils from the same vendor. The imaging protocol included a T1-weighted turbo spin-echo (T1 TSE) sequence (TR/TE = 1220/11 ms, acquisition matrix = 512×384,in plane resolution = 0.98×0.98 mm2, slice thickness = 5 mm, slice
gap= 5 mm, parallel imaging acceleration factor (iPat) = 2, 40slices, flipangle = 150◦, bandwidth = 219 Hz/pixel,turbofac- tor = 5, acquisition time = 2:15 min/40 slices, 160 slices by patients)anda T1-weighted TSEsequenceforapplyingthethree point Dixon method[30–33](TR/TE = 829/15.7 ms, acquisi- tion matrix = 512×384,in planeresolution = 0.98×0.98 mm2, slicethickness = 5 mm, slicegap= 5 mm,iPat = 2, 40slices, flip angle = 150◦, bandwidth = 315 Hz/pixel, turbo factor = 3, echo spacing = 15.7, acquisitiontime = 4:38 min/40 slices, 160 slices by patients). WaterandFatimages wereautomatically generated by the scanner from the TSE images for the three point Dixon method(Fig. 1). Bothsequences hadexactlythesameslice posi- tion and orientation. Image volume covered the proximal tibia to the lumbar spine (Th12 vertebra) andwasacquired in three stages. Total acquisitiontimewas45 min.
2.3. 3D muscle reconstruction(DPSO method)
The 3Dreconstruction ofindividual muscleswasperformed using Muscl’X software, a custom software, (Laboratory of Biomechanics, Arts et Métiers ParisTech, France). Using the axial MR images, the software generated the 3D geometry of each muscle. The reconstruction techniquewasbased on the DPSO algorithm as described in the literature [27–29]and briefly summarized hereafter(Fig. 2).
Foreachmuscle, a subset of MRI axial slices (MSS: manu- ally segmented slices) were manually segmented. The optimal percentage of MSS (defined asthe number of MSS divided by the total number of slices covering the entire muscle) of each muscle is reported in Table 1 [34]. Of note, complex muscle geometries required a larger percentage of MSS than simple geometries. Using contrast differences, these manual segmen- tations were then optimized (Fig. 2a). The contours were then approximated byellipses(Fig. 2b)andcubicsplineinterpolation wasusedtointerpolate ellipsesin all non-outlined slicescover- ingthe muscle(Fig. 2c). Theseinterpolated ellipses generated a 3D parametric object(Fig. 2d). Finally, usinga krigingalgorithm [35], the parametric objectwas deformed non-linearly using the manual segmentations of MSS as control points (Fig. 2e). Contrast enhancements were used to optimize the segmen- tation at each slice. Once all muscles were reconstructed, a geometry-correction algorithmwasapplied to eliminate mus- cle segmentation interpenetration and a visual verificationwas performed to correct local geometric deformities. Finally, 3D meshedreconstructions of each muscle were obtained.
Fig. 2. Overview ofthereconstruction method[28]:a:subset of MRIaxialslices with manually segmented axial sections; b: contours approximated by ellipses [represented on the figure by a green rectangle whose length and width corre- spondtothe major and minor axes ofthe ellipses]; c: cubic splineinterpolation usedtointerpolateellipsesinall non-outlinedslicescoveringthe muscle; d:inter- polated ellipses generated a 3D parametric object; e: non-linear deformation of the parametric objectwasdeformed usingthe manual segmentations.
2.4. Muscles ofinterest
Table 1describestheright- andleft-sided muscles or groups of musclesanalyzedinthisstudy. Muscles werechosen based on their potentialroleinregulatingthe position ofthe pelvisandthe spine. Because the delineation of certain muscleswasdifficult and in order to decrease the number of reconstructed muscles,
Table 1
Muscles analyzedinthis study, and optimal percentage of MSS slices. Musclereconstructed(right andleft) Percentage of MSS
Adductor 20
Bicepsfemoris 12
Erector spinae 15
Gluteus maximus 18
Gluteus medius 25
Gluteus minimus 30
Gracilis 10
Iliacus 25
Obliquus 20
Psoas 10
Quadratuslumborum 18
Rectus abdominus 12
Rectusfemoris 13
Sartorius 10
Semi-membranoustendinosis 11
TensorFascia Lata 15
Vastuslateralisinter 15
Vastusmedialis 15
MSS: manually segmented slices.
some ofthe muscles wereregrouped. The adductorlongus, bre- vis and magnus were reconstructed into a single group named
“Adductor”.In addition, the transversus abdominis muscle, internus obliquus, and externus obliquus were considered as a single group named “Obliquus” (they were reconstructed from their caudalinsertion uptotheliver). Therectus abdominus,the psoas, andthe erectorspinae werereconstructedfromtheir cau- dalinsertiontothesuperiorendplate ofthe firstlumbar vertebra. The short- andlong- heads ofthe biceps femoris were grouped together(“Bicepsfemoris”), as werethesemimembranosus and semitendinosus muscles(“Semi-membranoustendinosus”),and the vastuslateralis and vastusintermedius muscles(“vastuslat- eralisinter”).Intotal, 18 muscles,right andleft, were analyzed. 2.5. MR parameters
Based on 3Dreconstructions,thefollowing parameters were calculated for each muscle: mean and maximal axial section (AS), muscle volume, and muscle length. Muscle lengthwas defined asthe sum of distances between barycenters of all con- secutives slices.
From the Water andFatimages, the relativefatcontent for each voxel,fat/waterratio,wascalculated using Eq.(1).
Fat/Water Ratio=100× SIfat
SIfat+SIwater (1) where SIfatrepresentsthe signalintensity oftheFatimage and SIwaterrepresents the signal intensity of the Water image.For each muscle, all voxels contained in the muscle outline were identified. The average fat/water ratiowascalculatedoverall voxelscontainedineach muscleallowinga quantification ofthe fatinsideeach muscle.
2.6. Evaluation ofthe protocol
All the operators involved in the protocol evaluation were experiencedin reading muscular anatomy on MR images and receivedtraininginthe use ofthe software (Muscl’X).Forthe DPSO method, the first and last slices of each muscle were manually segmented and used aslimits for all reconstructions; all operators used exactly the same number of MSS (Table 1) whilethe selection ofthe actual MSSwaslefttothe operator’s discretion.
2.7. Evaluation ofthe DPSO method
Foreach muscle ofthestudied volunteers, areference object (i.e. 3D geometry)wasgenerated on T1 images by manually contouring all MR axialimages coveringit.Fig. 3presentsthe 3D reconstruction of volunteer A. This method of reconstruc- tionwasnamedthereference method andthe 3Dreconstruction objects were namedreferences. Thereferencesfor volunteers A and B were obtained by one operator. Usingthe DPSO method andthe number of MSS presentedinTable 1,this operatorrecon- structed once ontheFatimages and once onthe T1images all the musclesfor both volunteers.