Received: 15 July 2002Accepted: 11 November 2002© ESMRMB 2003

Abstract Embryonic developmentin normal and genetically modifiedmice is commonly analysed by histological sectioning. This proce-dure is time-consuming, prone toartefact, and results in the loss ofthree-dimensional (3D) information.Magnetic resonance imaging (MRI)of embryos has the potential of non-invasively acquiring a complete 3D data set. Published methods haveused spin-echo techniques with inherently high signal-to-noise ratio(SNR); however, they required either perfusion of the embryo witha contrast agent, or prolonged acqui-sition times to improve contrast and resolution. Here, we show that a standard preparation (i.e. parafor-maldehyde fixation) of 15.5 dayspost-coitum embryos followed byMRI using a fast gradient-echo sequence with T1-weighting

achieves high resolution and highthroughput for investigating mouseembryonic anatomy. 3D data setswere acquired in overnight experi-ments (<9 h) with an experimentalresolution of approximately 25 µm3.This spatial resolution is twofoldhigher than the values reported previously for comparable parafor-maldehyde-fixed embryos, and itwas obtained in less than a quarterof the time with sufficient SNR. Our approach combines speed, highresolution and contrast with a simplepreparation technique and minimaloperator time (<1 h). It allows rapidroutine 3D characterisation of normal and abnormal mouse embry-onic anatomy.

Keywords Magnetic resonance imaging · Mouse · Embryo · Histology · Foetal development

MAGMA (2003) 16:43–51DOI 10.1007/s10334-003-0002-z R E S E A R C H A RT I C L E

Jürgen E. SchneiderSimon D. BamforthStuart M. GrieveKieran ClarkeShoumo BhattacharyaStefan Neubauer

High-resolution, high-throughput magneticresonance imaging of mouse embryonic anatomy using a fast gradient-echo sequence

Introduction

Transgenic (gene-knockout) and high-throughput muta-genesis approaches are currently being used to rapidlygenerate large numbers of mutant mice in order to definethe genetic basis of human disease and development. Thephenotypic effects of a mutation range from physiologicalchanges without overt anatomical defects to severe devel-opmental malformations that are lethal in utero or soon af-ter birth [1]. These developmental malformations providemodels for many important human congenital diseases in-cluding neural-tube defects and cardiac malformations,which together affect almost 2% of all newborns [2, 3].Importantly, many major developmental phenotypes, such

as neural-tube defects, are significantly modified by varia-tion in the environment or at other gene loci [4]. Thesemodifier effects frequently cause a quantitative, ratherthan a qualitative change in the final phenotype. It is pos-sible to identify the modifying gene or environmental fac-tor provided the phenotype can be accurately measured[4]. Mapping and cloning genes mutated by chemical mu-tagenesis also requires accurate high-throughput pheno-typing capacity. Hence, there is an urgent need for facilemethods that allow not only rapid identification, but alsomeasurement of the severity of congenital defects in therapidly expanding numbers of murine models.

Traditional methods such as histological sectioningrequire considerable time for analysis [5]. Another

J. E. Schneider (✉) · S. D. BamforthS. Bhattacharya · S. NeubauerDepartment of Cardiovascular Medicine,John Radcliffe Hospital,University of Oxford,Headley Way, Oxford, OX3 9DU UKe-mail: jurgen.schneider@cardiov.ox.ac.ukTel.: +44-1865-282543Fax: +44-1865-22207

S. M. Grieve · K. ClarkeDepartment of Biochemistry,University of Oxford, Oxford, UK

major limitation of these methods is the deconstructionof a native three-dimensional (3D) object to a series of two-dimensional (2D) sections. The process of sectioning an embryo into thin slices of typically a fewmicrons also requires considerable skill and is prone to error due to mechanical deformations when cuttingthe embedded specimen. A trained technician with an optimal set-up needs at least 2 weeks of ‘hands-on-time’ for processing a single 15.5-day-old mouse embryo (body height approximately 15 mm) from head to tail until a digital data set is obtained. Cor-rection and subsequent reconstruction of the digi-talised 2D sections to a 3D data set is a complex andnon-trivial procedure [6–10]. Therefore, these consider-ations place severe constraints on the rate and costs ofhistology and are a barrier for high-throughput embryoanalyses.

High-resolution magnetic resonance imaging (MRI) isinherently non-invasive, 3D-capable and provides intrin-sic contrast mechanisms. The application of this tech-nique in mouse embryos has been successfully demon-strated [5, 11–16], using 3D spin-echo (SE) sequenceswith repetition times ≥200 ms and typically 1282 or 2562

phase-encoding steps. In these studies, contrast and reso-lution have been improved using either long repetitiontimes (resulting in total experimental times of approxi-mately 36 h for 15.5-day-old paraformaldehyde-fixedembryos), or the perfusion of the embryo with fixativeand MR contrast agent [12]. Perfusion of embryos involves catheterising the umbilical vein and thereforerequires operator skill and time. These skill and timeconstraints are relative disadvantages for the applicationof MRI to complement the high-throughput genetic mu-tagenesis strategies currently in progress.

Gradient-echo (GE) sequences are inherently faster,but more demanding for the spectrometer hardware (i.e.gradients) and have not yet been used for studying em-bryonic development. In this report, we show that a fastGE sequence achieves high-resolution images of the em-bryos and can potentially be used for high-throughputMRI of mouse embryonic anatomy. We imaged parafor-maldehyde-fixed normal mouse embryos at 15.5 dayspost-coitum in overnight experiments (<9 h) using a fastspoiled 3D GE sequence with T1-weighting. A matrixsize of 512×512×768 at a field of view (FOV) of13×13×20 mm3 led to an experimental resolution of ap-proximately 25 µm3. This spatial resolution is twofoldhigher than previously reported values for comparableparaformaldehyde-fixed embryos, and was obtained inless than a quarter of the time [17]. The signal-to-noiseratio (SNR) was sufficient to identify even small ana-tomical details. Our approach combines speed, high res-olution and contrast with a simple preparation techniqueand minimal operator time (<1 h), and has the potentialfor rapidly investigating normal and abnormal mouseembryonic anatomy.

Materials and methods

Animal preparation

C57Bl/6 mice were mated and the detection of a vaginal plug thefollowing morning was considered 0.5 days post-coitum (dpc).Pregnant female mice were sacrificed by cervical dislocation at 15.5 dpc, the embryos harvested and fixed in 4% paraformalde-hyde in phosphate-buffered saline (pH 7.4) at 4 °C until imaged.For MRI experiments, embryos were embedded in a 1% agarosegel containing 2 mM Gd-DTPA (Magnevist, Schering AG, Burgess Hill, West Sussex, UK) in 13-mm diameter NMR tubes.In total, five normal embryos were examined.

Magnetic resonance set-up

Experiments were carried out on a 11.7-Tesla (500 MHz) verticalmagnet (Magnex Scientific, Oxon, UK) interfaced to a BrukerAvance spectrometer (Bruker Medical, Ettlingen, Germany) andequipped with a shielded gradient system (548 mTesla/m, rise time160 µs) (Magnex Scientific). A linear-driven birdcage coil with an inner diameter of 13 mm (Rapid Biomedical, Würzburg, Germany) was used to transmit/receive the NMR signals. Com-pressed air (flow: 2,000 l/h) at room temperature was used to compensate for gradient-induced heating. Tuning and matchingthe probe, global shimming and flip-angle calibration were donemanually prior to each experiment.

Contrast optimisation

For contrast optimisation, 3D spoiled GE sequences with a matrixsize of 256×256×512 at a FOV of 13×13×20 mm3 were applied.Different degrees of T1-weighting were achieved by varying theflip angle (16°, 45° and 90°) at a constant repetition time ofTR=30 ms. Data sets with echo times of TE=5 ms, 10 ms and15 ms, respectively, were acquired to investigate T2

* contrast.These data sets were also used for a quantitative T2

*-estimation ofembryonic tissue.

Anatomical imaging

High-resolution MRI of embryonic anatomy was carried out usingthe 3D spoiled GE sequence with an echo time of TE=10 ms. A90° excitation pulse (rectangular pulse shape, 90°=100 µs) and ashort repetition time (TR=30 ms) resulted in strong T1-contrast inthe images. A matrix size of 512×512×768 (frequency encoding inz-direction; bandwidth: 130 Hz/pixel) at a FOV of 13×13×20 mm3

led to a nearly isotropic resolution of 25.4×25.4×26.0 µm3. The total experimental time was less than 9 h, whereby each phase-encoding step was averaged four times (NA= 4).

Data analysis and segmentation

The raw MRI data sets (with a size of up to more than 1.6 GB)were reconstructed on a 1.4-GHz Linux-PC using a purpose-written C-software. Data of the contrast study were zero-filled bya factor of two in each direction, whereas the high-resolution datawere isotropically zero-filled to 1,024 points in each dimension. A modified Butterworth filter function [18] was applied alongeach axis prior to Fourier transformation.

For T2*-estimation, slices through the head, the chest and the

abdomen were extracted from the 3D data set and fitted using the idl-software (Research Systems International, Crowthorne,Berkshire, UK). For advanced visualisation and image analysis of

44

the anatomical data, axial 2D TIFF-images were generated fromthe 3D data set with an 8-bit or 16-bit resolution per pixel. The final processing was then done with Amira 2.3 (TGS Europe,Mérignac Cedex, France), which provided tools for segmentation,direct volume rendering and computation of isosurfaces. Struc-tures of interest were automatically or semi-automatically outlinedin the 2D sections using the threshold detection “magic-wand”tool or the auto-trace function in Amira. These data were used toconstruct 3D renderings of each embryo that could be rotated inspace and viewed along arbitrary planes using the Amira software.Anatomical structures were assigned according to [19].

To assess reproducibility of our method, body volumes as wellas kidneys and adrenal glands of all five embryos were segmented.The accuracy of quantitative results was determined in a separatephantom experiment using a pear-shaped glass sphere of a knownvolume filled with distilled water.

Results

An MRI protocol designed to compete with traditionalhistology, which has intrinsically high spatial resolutionand multiple sources of contrast, must maximize nativeMR properties in order to provide comparable anatomi-cal information. In practice, maximising the anatomicaldetail requires a compromise between contrast, spatialresolution and experimental time. To establish an opti-mal experimental configuration, various data sets were

acquired using a 3D spoiled GE sequence over a range ofdifferent excitation angles and TE values. In Fig. 1 examples of slices through the head, the chest and theabdomen are shown. They were acquired with a flip an-gle of 90° at echo times of 5 ms, 10 ms and 15 ms, re-spectively. An increasing echo time led to an enhance-ment of tissue interfaces facilitating the identification ofdifferent organs. The overall (non-tissue-specific) con-trast due to variation in T1-weighting and diamagneticsusceptibility was found to be optimal for a combinationof high flip-angle/short TR and an intermediate TE (Fig. 1D–F). Accordingly, a 90° excitation pulse at aTR=30 ms with a TE=10 ms was chosen for high-resolu-tion imaging of embryonic anatomy.

The same data were also used to estimate embryonictissue values of T2

*, which ranged between 5 ms and40 ms (corresponding line-width between 8 Hz and63 Hz). The full line-width at half maximum (FWHM)after global shimming – dominated by the Gd-DTPA-containing compound of the agarose – was typicallyabout 80 Hz (0.16 ppm) and therefore still well belowthe used bandwidth of 130 Hz/pixel.

Figures 2 and 3 show coronal and axial views from a3D high-resolution MRI data set of a 15.5 dpc murineembryo using the sequence parameters described above.

45

Fig. 1 A–I Axial sectionsthrough head, chest and abdo-men obtained by MRI from afixed mouse embryo at15.5 days post-coitum using a3D spoiled gradient-echo (GE)sequence with a 90° excitationpulse, TR=30 ms and differentecho times. Tissue interfaces areenhanced with increasing TE

The pixel size in Figs. 2 and 3 is 13×20 µm2 and13×13 µm2, with a “slice thickness” of 13 µm and 20 µm,respectively. The high-resolution MRI data set was ac-quired within approximately 8.5 h, and about 1.5 h wereneeded for reconstruction and generation of the TIFF im-ages, within which handling of the large data file (morethan 8 GB after Fourier transformation in all dimensions)was the most time-consuming component. Major organssuch as lungs, liver, kidneys or stomach could easily beidentified (Figs. 2 and 3). The combination of contrastand resolution obtained allowed the detection and identi-fication of small anatomical structures such as the pitui-

tary and adrenal glands (Figs. 2B, 3D), as well as thecardiac chambers, atrial and ventricular septa, and greatvessels (Figs. 2 and 3).

Differences in contrast also facilitated segmentationand quantitative analysis of various organs or tissuetypes. Paramagnetic contrast agent (Gd-DTPA) in agaro-se combined with T1-weighting enabled fully automaticsegmentation of the whole embryo out of the surround-ing agarose matrix, as shown in Fig. 4A. Inside the embryo, structures such as brain and spinal cord (Fig. 4B) or kidneys with adrenal glands (Fig. 4C) couldeasily be segmented using a combination of automaticand manual segmentation.

In Fig. 5, different views of a segmented embryonicheart are shown. Left and right ventricles, atria and greatvessels were assigned separately. In Fig. 5F, the ventri-cles and the outflow tracts are shown. The communica-tion between the pulmonary artery and the aorta (ductusarteriosus) is clearly visualised after digitally removingthe overlying structures.

An experienced user required approximately 10 h for setting up the MRI experiment, starting the recon-struction and analysing the entire embryonic data set (i.e. segmentation of different structures) as describedabove. The total analysis time per embryo (without fixa-tion) was therefore about 20 h.

Segmentation of the pear-shaped glass sphere filled257 µl of distilled water yielded 224 µl, which is 87.2%of the “true” volume. The segmented volumes for all fiveembryos are shown in Fig. 6A and compared to theirbody weight. The correlation coefficient was 0.94. Forthe averaged density (i.e. body weight/segmented vol-ume) we obtained (1.067±0.017)×103 kg/m3 (mean±SD).In Fig 6B, C, the volumes of the kidneys and adrenalglands normalised to the segmented body volumes areshown for the left and right sides separately. The follow-ing mean normalised volumes were obtained(mean±SD): left kidney (2.94±0.20)×10-3, right kidney(3.11±0.30)×10-3, left adrenal gland (3.06±0.13)×10-4,right adrenal gland (2.80±0.29)×10-4, resulting in a vari-ation between 4 and 10% for the segmented organs.

Discussion

There is a major demand for a fast and largely user-inde-pendent technique to analyse the growing number ofgene-manipulated developmental mutant mouse models.Designing an efficient method for routine use requiresminimising the time consumption per specimen in orderto increase the total throughput. Of particular interest arerobustness and reliability of the procedure. Traditionalhistological sectioning is the only method in widespreaduse for analysing the embryonic development of geneti-cally modified animals. However, this technique is limited by complex and time-consuming preparation and

46

Fig. 2 A–C Coronal sections obtained by MRI from a fixed mouseembryo at 15.5 days post-coitum. Scale bar 1 mm. D Midline sagittal section of the embryo indicating the section planes. l-lungLeft lung, V trigeminal (Vth) nerve ganglion, np nasopharynx,rvot right ventricular outflow tract, ivs interventricular septum,lv left ventricle, ra right atrium, rv right ventricle, d diaphragm

processing required per specimen. Hence, high-through-put screening projects are extremely costly and nearlyimpractical.

MRI provides intrinsic contrast and 3D capability anddoes not require a lengthy preparation procedure. A keylimitation of this technique is the achievable resolutionwithin acceptable times. Given a fixed experimental timeand optimised experimental design (i.e. hardware, meth-od), the spatial resolution of an MRI experiment is thenmainly limited by the physical properties of the sample.The NMR signal available from a volume element isproportional to the voxel volume, so the SNR is the mostsignificant factor determining the achievable resolution.Given an isotropic voxel with homogeneous spin distri-bution, the SNR scales with the third power of the voxellength. Averaging the MR experiment N times increasesthe SNR by a factor of . It follows that the experi-mental time scales inversely with the sixth power of vox-el length [20]. Further factors limiting the resolution arespin-spin relaxation, susceptibility, and diffusion [21].

In previous studies, high-resolution 3D MRI data setsof mouse embryos that were perfused with fixative andMR contrast agent were obtained using a SE sequencewith short echo times and T1-weighting [5, 11–13]. Ded-icated RF coils with a crown–rump length ranging frombetween 3 mm to almost 30 mm – depending on the size

of the embryo – and a matrix size of 2563 resulted in acalculated resolution of 12–120 µm, obtained within7–14 h of imaging time. Based on this technique, cono-truncal heart development of connexin 43 (Cx43) knock-out mice and CMV43 transgenic mice has been investi-gated [14]. Alternatively, the application of a spin-lockpulse train for T1ρ-contrast in MR histology of perfu-sion-fixed mouse embryos has been reported [15].

Spin-echo vs. gradient-echo

The purpose of our study was to establish high-resolu-tion MRI as a routine tool for analysing embryos of genetically manipulated mice. The emphasis was to ac-quire an entire 3D data set overnight, when an MRsystem is typically not required for user-interactive stud-ies, and to keep the preparation of the embryos as simpleas possible. To establish this technique, 15.5-dpc-oldnormal mouse embryos were chosen, since their cardiacand other organ development is complete at this stage.Prior to MR experiments, the only requirement was tosubject the embryos to fixing in a standard fixative solu-tion (paraformaldehyde). For MRI, a spoiled 3D GE waschosen rather than a SE sequence. SE sequences providedifferent contrast mechanisms compared to GE-sequenc-

47

Fig. 3 A–D Transverse sectionsobtained by MRI from a fixed mouse embryo at15.5 days post-coitum.Scale bar 1 mm. E Midlinesagittal section of the embryoindicating the section planes.ao Aorta, l-svc left superior vena cava, la left atrium,laa left atrial appendage, lv leftventricle, ivs interventricularseptum, rv right ventricle,ra right atrium, r-svc right su-perior vena cava, ias interatrialseptum, es esophagus, drg dor-sal root ganglion, lk left kidney,s stomach, int intestine, rk rightkidney, rad right adrenal

es as well as inherently high SNR. However, 512×512phase-encoding steps, as used in our protocol, and a repetition time of 200 ms for an SE as suggested bySmith [12], would result in a total experimental time ofapproximately 14.5 h without averaging. Engelhardt andJohnson measured T2 values of diencephalon, muscleand liver tissue of a 17.5-dpc-old perfusion fixed embryoat 2 Tesla and 9.4 Tesla [15]. Based on their investiga-tions, intrinsic T2-contrast at 12 Tesla would require TEvalues in an SE sequence to range between 30 and 50 msand therefore lead to an increased repetition time com-pared to our experiments. In particular, T2-weighted datasets at 11.7 Tesla were acquired for the digital mouse at-

las using TE values in a range of 25 ms and 200 ms incombination with repetition times of 1,750 ms and3,900 ms [17]. According to these considerations, the ac-quisition of a high-resolution 3D data set with 5122

phase-encoding steps, as applied, may not be possible inan overnight run. GE sequences used for high-resolutionMRI are, on the other hand, more demanding for thespectrometer hardware. Additional parameters such asgradient-induced heating or duty cycle of the gradientsystem have to be considered for a successful, unattend-ed overnight experiment.

Signal-to-noise ratio, resolution and contrast

To develop our method, sequence parameters of the GEwere varied to investigate different contrast mechanisms.It was observed that most of the sample contrast wasprovided by a combination of proton density and T1.Small variations of diamagnetic susceptibility led to an edge enhancement at the tissue interfaces. This facili-tated particularly the distinction and identification of or-gans in the abdominal area. Accordingly, a 90° excita-tion pulse combined with a short repetition time of30 ms and an echo time of 10 ms were applied. It is im-portant to note that the parameters were not optimisedfor a particular tissue type, but resulted in excellent con-trast within the entire embryo. SNR and contrast weresufficient to identify also small structures such as the pituitary and adrenal glands or the abdominal aorta.

In order to reduce the influence of radiation dampingon line shape and line width, shimming was performedwith high flip-angles and short repetition times. There-fore, the global FWHM of about 80 Hz (correspondingto a T2*≈4 ms) reflected mainly the paramagnetic com-pound in the agarose with short T1 and T2 values ratherthan tissue water contributions. A shift in resonance frequency of this compartment due to an interaction withthe paramagnetic gadolinium could also lead to a broad-ening of the enveloping line. To assess the distribution of T2

* values locally, T2* was estimated based on the

contrast experiments. The values were in the range of5–40 ms (corresponding line-width between 8 Hz and63 Hz) within the embryo. These results confirmed thatlocal T2

* values inside the volume of interest were betterthan expressed by the global line-width after shimmingand particularly well below the chosen bandwidth of130 Hz/pixel. Filtering the raw data prior to Fouriertransform not only enhanced SNR, but was also requireddue to zero-filling. The chosen modified Butterworth fil-ter, however, led only to a broadening of about 20%compared to the unfiltered case.

For anatomical imaging, an echo time of 10 ms waschosen, as it provided an acceptable compromise be-tween signal loss due to diamagnetic susceptibility andenhancement of tissue interfaces. As an additional ad-

48

Fig. 4 A Reconstruction of external embryonic surface. Segmen-tation was performed on transverse MRI sections obtained from asecond 15.5 days post-coitum mouse embryo using Amira 2.3.The 3D reconstructions were computed using the General March-ing Cubes algorithm. The red patches indicate the areas of contactwith the NMR tube. B Reconstruction of the external surface ofthe brain and spinal cord. C Reconstruction of the kidney and ad-renal glands. med Medulla, tel telencephalon

vantage, the longer TE value allowed increased phase-encoding and read-dephasing times and therefore reduc-tion of signal losses induced by gradient diffusionweighting as well as gradient-induced heating of thesample.

Embryo preparation

In spite of the simple embryo preparation, the contrastand resolution achieved by this protocol allowed the detection and identification of anatomical structures. Importantly, perfusion of the embryo through the umbili-cal vein, which is technically difficult and time-consum-ing, could be avoided. As the embryos were not per-fused, in occasional embryos we observed the presenceof blood in the cardiac chambers or vessels. The in-creased susceptibility caused by the paramagnetic iron inblood clots may result in signal loss and reduce the reso-lution locally. These local signal voids may potentiallyhamper the segmentation of the ventricular cavities andmay influence the quantitative analysis. However, in ourexperience, this does not interfere with a qualitative seg-mentation, as demonstrated in Fig. 5. Based on the sim-ple preparation, investigation of embryos at later stage isalso possible (data not shown). To compensate for re-

duced skin permeability as observed in older embryos,the same contrast can be obtained by carefully slittingthe skin of the embryo at both sides before fixation.

The size (diameter) of the NMR tube has to be adapt-ed to the respective embryo size (volume) for optimalsensitivity of the MR experiment. Any volume outsidethe embryo is redundant but is required in this set-up toimmobilise the embryo. We found that a NMR tube withan outer diameter of 13 mm is a good compromise between the optimal filling factor for the examined 15.5-dpc embryos and the practicability of preparation.Such a set-up also minimises the existence of any un-wanted signal (i.e. from the surrounding medium). Doping the agarose amplified the “unwanted” signal andgenerated a bright contrast from the surrounding mediumwhich was required as the method is based on a combi-nation of T1 and T2

* contrast (both resulting in signal reduction). This could be characterised as “negative contrast” – tissue and organs were visualised and sepa-rated by signal reduction and not by signal increase. Particularly, using doped agarose to embed the embryosresulted in a clear agarose/skin interface that allowed aneasy and fast segmentation of the entire embryo. Howev-er, this approach emphasises that the ratio between tubevolume and embryo volume has to be optimised in orderto obtain the best sensitivity for the embryo.

49

Fig. 5A–F Reconstruction ofthe heart and great vessels ofthe 15.5 days post-coitummouse embryo shown in Fig. 3.The internal outlines of cardiacchambers and great vesselswere identified and marked,first in transverse and then insagittal MRI sections, usingAmira 2.3. The 3D reconstruc-tions were computed using theGeneral Marching Cubes algo-rithm. A Right lateral view,B anterior superior oblique,C left lateral, D posterior-later-al, E anterior-inferior oblique,F as in E, with great veins andatria removed to show the ven-tricles and their outflow tracts.r-svc Right superior vena cava,ra right atrium, rv right ventri-cle, lv left ventricle, ivc inferiorvena cava, ao aorta, l-svc leftsuperior vena cava, pv pulmo-nary veins, pa-ductus pulmona-ry artery and ductus arteriosus

Segmentation and 3D reconstruction

As both normal and malformed hearts are complex 3Dstructures, the ability to use segmentation analysis to reconstruct the heart in 3D vastly enhances our ability tounderstand foetal heart topography and to accuratelyquantify malformations.

Segmenting various structures of the embryo demon-strated the capacity of the method as a powerful tool toanalyse a broad range of organs in the mouse embryo.Most of the segmentation was automatically done usingthe tool “magic-wand” which allowed application of alower and upper threshold symmetrically around thecontrast of a specified pixel. If isolated pixel needed tobe added to a tissue type, a manual correction was ap-plied (semi-automatic segmentation). Both the kidneys(inhomogeneous tissue-type) and the adrenal glandswere segmented using the auto-trace tool, which allowedthe tissue borders to be followed automatically. This pro-cedure again was facilitated by the increased T2

* contrastthat visualised tissue interfaces, particularly in the ab-dominal region, more clearly.

Based on image segmentation, areas and volumes ofvessels and organs can be calculated. The fixative (PFA,Bouins, etc.) and the chosen MR sequence need to beconsidered as they may influence any quantitative result.To validate the quantitative analysis of the embryonicdata sets, an NMR experiment (same timing/resolution)on a phantom with known volume rather than an histo-logical procedure was chosen. The preparation for histol-ogy, i.e. dehydration, results in substantial shrinking oftissue and therefore adulterates any subsequently calcu-lated organ/tissue volumes considerably. We yielded anunderestimation of the “true” volume of about 13% inour phantom experiment. This is not surprising as thereis signal loss at the glass-water interface during the rela-tively long echo time of 10 ms due to a substantial varia-tion of diamagnetic susceptibility. However, this effect islikely to be of smaller magnitude in the fixed embryos.We also found the longer echo time to be favourable as itfacilitated segmentation, particularly in abdominal area,as mentioned above.

The segmented volumes of the entire embryo bodiescorrelated well with their body weight. For determiningthe body volumes, the surrounding agarose-matrix wassegmented first, using the “magic-wand”-tool. The re-maining volume within the matrix was then assigned asembryonic tissue. An underestimation of the agarose volume would result in an overestimation of the embryovolume and therefore decrease the calculated density.However, a mean factor of (1.067±0.017)×103 kg/m3

(mean±SD) agrees well with values reported for tissue invivo (e.g. [22]). The variation in body volumes (weights)is obviously dominated by anatomical and temporal vari-ation, since mating may occur between 15 and 16 daysbefore harvesting the embryos at the stage of 15.5 dpc.

50

Fig. 6 A Segmented body volumes (◆, in µl) and weight (●, inmg). The correlation coefficient was 0.94. B Volumes of left (◆)and right (■) kidneys. C Volumes of left (▲) and right (×) adrenalglands. The volumes shown in B and C were normalised to thesegmented body-volume. The mean values were (mean±SD): leftkidney (2.94±0.20)×10−3, right kidney (3.11±0.30)×10−3, left adre-nal gland (3.06±0.13)×10−4, right adrenal gland (2.80±0.29)×10−4

Accordingly, volumes of kidneys and adrenal glandswere normalised to their body volume to correct forthese variations. For these tissue types, normalised vol-umes were found to vary between 4 and 10%, demon-strating the high reproducibility of our method. The vari-ation is likely to be dominated by the limited accuracy ofsegmentation and by anatomical differences. The resultsindicate that quantitative information can be expectedwithin an accuracy of 10%, which is sufficient to quanti-tatively identify abnormalities in embryonic develop-ment.

Total time for microscopic analysis of one embryo in-cluding high-resolution MRI and advanced post-process-ing was approximately 20 h, only half of which involveduser input. In most cases, segmentation of the data setmay not be necessary, as standard analysis involvingscrolling through slices and qualitatively describing ana-tomical variants may be sufficient to characterise embry-onic malformations. Such a qualitative approach wouldresult in an additional reduction of the post-processingtime. In the future, the use of a birdcage coil driven inquadrature mode would increase the SNR by a (theoreti-cal) factor of or halve the experimental time with thesame SNR [23]. Thus, an additional gain in throughputmay be achievable.

Conclusion

A MR method for investigating mouse foetal develop-ment has been implemented with an experimental resolu-tion of approximately 25 µm3 and excellent contrast be-tween different tissue types. Anatomical details could bevisualised and were segmented using image post-processing software. Quantitative analysis of segmenteddata showed excellent reproducibility with an accuracybetter than 10%. For a method that does not requirelengthy and artefact-prone preparation of the embryo byperfusion with fixing solution prior to the MRI experi-ment, our approach achieves the highest resolution at theshortest measurement time reported to date. Moreover, itrepresents a substantial improvement over the time re-quired for histological preparation and analysis of em-bryos and is suitable for routine high-level throughput.

Acknowledgements This work was supported by the BritishHeart Foundation and the Wellcome Trust. We thank Dr GraemeWaddington for helpful discussions.

51

References

1. Chien KR (1996) Genes and physiolo-gy: molecular physiology in genetical-ly engineered animals. J Clin Invest97:901–909

2. Gardner-Medwin D (1986) Develop-mental abnormalities of the nervoussystem. Oxford textbook of medicine.Oxford University Press, Oxford, pp 21190–21206

3. Clark EB (2001) Etiology of congenitalcardiac malformations: epidemiologyand genetics. In: Driscoll DJ (ed).Moss and Adams’ heart disease in infants, children and adolescents. Lippincott Williams and Wilkins, Philadelphia pp 64–79

4. Juriloff DM, Harris MJ (2000) Mousemodels for neural tube closure defects.Hum Mol Genet 9:993–1000

5. Smith BR, Linney E, Huff DS, Johnson GA (1996) Magnetic reso-nance microscopy of embryos. ComputMed Imaging Graph 20:483–490

6. Toga AW (1998) Brain warping. Academic , New York

7. Toga AW, Ambach KL, Quinn B,Shankar K, Schluender S (1996) Post mortem anatomy. In: Toga AW,Mazziotta JC (eds). Brain mapping: the methods. Academic, New York, pp 169–190

8. Toh MY, Falk RB, Main JS (1996) Interactive brain atlas with the VisibleHuman Project data: developmentmethods and techniques. Radiograph-ics 16:1201–1206

9. Brune RM, Bard JB, Dubreuil C, Guest E, Hill W, Kaufman M, Stark M,Davidson D, Baldock RA (1999) Athree-dimensional model of the mouseat embryonic day 9. Dev Biol216:457–468

10. Kaufman MH, Brune RM, DavidsonDR, Baldock RA (1998) Computer-generated three-dimensional recon-structions of serially sectioned mouseembryos. J Anat 193:323–336

11. Smith BR (2001) Magnetic resonancemicroscopy in cardiac development.Microsc Res Tech 52:323–330

12. Smith BR (2000) Magnetic resonanceimaging analysis of embryos. In: Tuan R, Lo CW (eds) Developmentalbiology protocols. Humana, Totawa,New Jersey pp 211–216

13. Smith BR, Johnson GA, Groman EV,Linney E (1994) Magnetic resonancemicroscopy of mouse embryos. ProcNatl Acad Sci USA 91:3530–3353

14. Huang GY, Wessels A, Smith BR, Linask KK, Ewart JL, Lo CW (1998)Alteration in connexin 43 gap junctiongene dosage impairs conotruncal heartdevelopment. Dev Biol 198:32–44

15. Engelhardt RT, Johnson GA (1996) T1rho relaxation and its application to MRhistology. Magn Reson Med 35:781–786

16. Jacobs RE, Ahrens ET, Dickinson ME,Laidlaw D (1999) Towards a micro-MRI atlas of mouse development.Comput Med Imaging Graph 23:15–24

17. Dhenain M, Ruffins SW, Jacobs RE(2001) Three-dimensional digitalmouse atlas using high-resolution MRI.Dev Biol 232:458–470

18. Gonzalez RC, Wintz P (1983) Digitalimage processing. Addison-Wesley,Reading Massachusetts

19. Kaufman MH (1994) The atlas ofmouse development. Academic, London

20. Mansfield P, Morris PK (1982) NMRimaging in biomedicine. Academic,New York

21. Callaghan PT (2001) Principles ofmagnetic resonance microscopy. Oxford University Press, Oxford

22. Manning WJ, Wei JY, Katz SE, LitwinSE, Douglas PS (1994) In vivo assess-ment of LV mass in mice using high-frequency cardiac ultrasound: necropsyvalidation. Am J Physiol266:H1672–H1675

23. Chen CN, Hoult DI, Sank VJ (1983)Quadrature detection coils – a further√2 improvement in sensitivity. J MagnReson 54:324–327