New Developments and Applications of the MP2RAGE Sequence - Focusing the Contrast and High Spatial Resolution R1 Mapping

MR structural T1-weighted imaging using high field systems (>3T) is severely hampered by the existing large transmit field inhomogeneities. New sequences have been developed to better cope with such nuisances. In this work we show the potential of a recently proposed sequence, the MP2RAGE, to obtain improved grey white matter contrast with respect to conventional T1-w protocols, allowing for a better visualization of thalamic nuclei and different white matter bundles in the brain stem. Furthermore, the possibility to obtain high spatial resolution (0.65 mm isotropic) R1 maps fully independent of the transmit field inhomogeneities in clinical acceptable time is demonstrated. In this high resolution R1 maps it was possible to clearly observe varying properties of cortical grey matter throughout the cortex and observe different hippocampus fields with variations of intensity that correlate with known myelin concentration variations.


Introduction
The promise of ultra high fields systems to provide higher spatial resolution structural images due to their higher signal to noise ratio (SNR) has been hampered by the increase in transmit magnetic field inhomogeneities and the high specific absorption rate (SAR) that come associated with the decrease of the wavelength and increased frequency of the RF pulses that are used to excite the proton spins [1]. SAR limitations have made gradient echo based sequences to be the main workhorse of structural imaging at high fields, delivering eitherT 2 * -weighted imaging [2], phase imaging [3] and T 1 -weighted imaging. The MP2RAGE sequence has been recently introduced as a means to obtain bias field free T 1 -weighted images and jointly estimating T 1 maps at ultrahigh field [4,5]. In the original work, the sequence parameter optimization was developed to obtain the conventional range of contrast used in T 1 -weighted imaging (covering the T 1 range from white matter to cerebro spinal fluid) with the protocols being defined in order to achieve reliable T 1 maps when the typical clinical whole brain isotropic resolution of approximately 1 mm was desired. Although such a large range of T 1 range and conservative resolution is desirable for normal brain imaging and segmentation applications, it is not ideal when looking at detailed visualization of deep gray matter structures [6] or fine variations in cortical relaxation properties throughout the brain [7].
Visualization of deep gray matter structures such as the different thalamus nuclei, striatum, external and internal globus pallidus (GPe/GPi), red nucleus (RN) and substantia nigra (SN) can be of great importance in applications such as deep brain stimulation (DBS) used in the treatment of involuntary movement disorders, in Parkinson's disease or dystonia [8,9]. Attempts to automatically segment the thalamus nuclei using structural imaging have to date had some success [10,11], but only limited correlation was found with thalamic segmentation using either diffusion weighted imaging [12] or histology. Due to the low levels of contrast between the basal ganglia and surrounding structures in T 1 -w images, a number of approaches using other contrast have been explored [10,[13][14][15]. Among these, a recent studies have proposed a modification of the standard MPRAGE sequence parameters to yield a better visualisation of basal ganglia structures [16] and subthalamic nuclei [17] by using a shorter inversion time than the usually used at the respective field strengths (in which the CSF signal is nulled). Recently in a 3T study, the MP2RAGE was modified [18] in order to provide two T 1 -weighted images, one with the conventional MPRAGE contrast (with CSF nulling) and one with a short TI where the WM signal is suppressed.
Despite the optimization of the MP2RAGE sequence parameters [5] performed in order to reduce B 1 + dependence of those images (at a cost of a reduction of the contrast obtainable), the resulting T 1 -maps still suffer from some residual transmit field bias. Furthermore, the temptation to increase the resolution (by increasing the number of low flip angle excitations per TR) and the need to keep the total acquisition time low (by reducing the TR of the MP2RAGE), increase the sensitivity of the MP2RAGE T 1 estimation to B 1 + inhomogeneities. Recently, some attention has been drawn to the correlation between the observed cortical T 1 values and known distributions of myelination [7,19,20]. For these correlations to be further evaluated at such high resolution and through such a large brain extent, it is imperative to obtain high resolution, robust and fully bias free T 1 values. Often, the effect of T 1 relaxation and B 1 + inhomogeneity on signal are intertwined and most techniques proposed to measure T 1 effectively measure these T 1 and B 1 + simultaneously [21][22][23][24][25][26]. This can be done either within one single acquisition, which impacts on the SNR efficiency of the method, or using separate acquisitions.
In this work, we explore different potential high field applications of the MP2RAGE sequence: (i) improved contrast between white matter and grey matter tissues allowing clear visualization of sub-thalamic nuclei and other deep grey matter structures; (ii) high resolution T 1 maps with optimum contrast to noise ratio in which the effects of significant B 1 field inhomogeneities are taken into account by combining the MP2RAGE sequence with the Sa2RAGE sequence [27] and study R 1 variations throughout the cortex and in different white matter bundles.

Theory and Methods
The MP2RAGE sequence can be described as an inversion recovery sequence in which two gradient echo images are acquired during the recovery period with different inversion times (GRE TI1 and GRE TI2 ). The MP2RAGE image is a synthetic image obtained from a combination of the two acquisitions: both images are taken as complex images and the asterisk stands for the complex conjugate.
The Sa2RAGE sequence is a saturation recovery sequence in which two images are acquired before (GRE TD2 ) and after (GRE TD1 ) a saturation pulse (see Figure 1 for a more complete description). A full Bloch equation simulation of the sequence permits the mapping of the transmit field B 1 . It was shown [27] that this technique could provide B 1 maps at 7T with small errors even when assuming a common T 1 value for the whole brain.

Simulations
The optimum sequence parameters for the different applications were studied via simulations. The predicted MP2RAGE signal amplitudes for several tissues were numerically calculated after solving the Bloch Equations as in reference [5].
When studying the (i) optimum contrast between white matter WM, thalamus and Gray Matter the following parameters were considered fixed: number of excitations per GRE module was set to 160 (full k-space coverage) or 120 (partial fourier k-space coverage); The contrast at 7 Tesla was evaluated for 5 different T 1 values, ranging from 1.1 (,T 1WM ) to 1.9s(,T 1GM ) [5]. The range of T 1 values chosen should cover the different T 1 values in the thalamus and the choice of the number of excitations was done to allow a 1mm isotropic resolution with full brain coverage in sagittal orientation.
When studying the (ii) optimum contrast achievable at high resolution (0.65 mm isotropic): Number of excitations per GRE module was set to 192 (partial fourier k-space coverage 6/8-256 partitions); 5 T 1 values equally spaced ranging from 1.1 (,T 1WM ) to 4 s (,T 1CSF ). The range of T 1 values was chosen in order to include all brain tissues and CSF, while the choice of the number of excitations was done to achieve a 0.65 mm isotropic resolution with full brain coverage in sagittal orientation. In both cases, the following parameters were varied in order to compute the optimum protocols: MP2RAGE TR , TI 1 and TI 2 , a 1 and a 2 (see Figure 1 for a complete overview of the sequence and the meaning of the different parameters).
Contrast to noise by unit of time between two tissues (i and j) with successive values of T 1 was defined as: Where S i is the calculated MP2RAGE signal [5] of tissue i, and its noise, s Si , was estimated by error propagation of the MP2RAGE signal equation. The contrast to noise ratio between the successive T 1 values was integrated in order to obtain the sequence CNR for the desired application.
Sequence parameters were chosen from the simulations in order to optimize the CNR for the desired T 1 range and for the desired resolution. A ground truth T 1 mapping protocol was designed to evaluate the accuracy of the T 1 correction methodology (experiment ii). The ground truth protocol consisted of an MP2RAGE acquisition with a reduced number of rf pulses during the recovery process (hence, to keep the FOV the same, a lower resolution was used) and reduced flip angle amplitude for both echo trains (ensuring the low B 1 + inhomogeneity sensitivity). In this protocol, the maximum achievable CNR was penalized in order to have a bias in the T 1 mapping under 5% for all tissues in the brain when in the presence of B 1 field deviations as large as 640% of the nominal value.

Experimental Protocol
Data were collected at a short-bore 7T MR system (Siemens, Germany) equipped with a head-gradient insert and a 32-channel head coil (Nova Medical Inc) for reception. All subjects provided written informed consent and the study was approved by the local ethics committee (Commission cantonale VD d'éthique de la recherche sur l'être humain).
Both acquisitions were performed using iPAT PE = 2 and 6/8 kspace coverage on the slice encoding direction, acquisition time of 10 mins. The matrix size and resolution were of 25662006176 and 0.85 mm isotropic respectively, while the BW was of 240 Hz per pixel.

Processing Protocol (i)
When performing the optimization of the MP2RAGE sequence parameters to enhance the contrast for a limited range of T 1 values (from 1.1 to 1.9 s), it is possible to observe (see Fig. 2a), that the relationship between T 1 and MP2RAGE signal intensity is no longer monotonous. The CSF intensity appears aliased to intensity values close to that of WM. Using the Bloch simulations it is possible to recover the full range by using information from a complex ratio between the second and first inversion time images [4], which is monotonic as a function of T 1 . Although this new image, MP2RAGE WMGMFullRange , has exactly the same contrast properties as the MP2RAGE WMGM , the fact that CSF appears dark makes the interpretation of the images easier.
Using the sequence parameters suggested to enhance subthalamic contrast, the intensity of WM and of CSF are close to zero in the first and second inversion contrasts respectively (note that in Fig. 2a both have an MP2RAGE intensity of ,0). This is a similar scenario to that observed in the FLAWS sequence, where this feature was used to create Double Inversion Recovery like images [18]. Because of the specificities of the receive and transmit field inhomogeneities observed at 7T, the MP2RAGE MIP (minimum intensity projection) images were computed as: To obtain a quantitative evaluation of the contrast obtained using the different sequence parameters, regions of interest were defined on the images of protocol A (MP2RAGE) and protocol B (MP2RAGE WMGM ) using MRIcro [30] and their intersection was used to evaluate the mean and standard deviation of the signal. The contrast to noise ratio between different structures was quantified as in Eq. 2.

Processing Protocol (ii)
The Sa2RAGE image (ii-c) and the low resolution MP2RAGE image (ii-b) were co-registered to the high resolution MP2RAGE image using FLIRT (www.fmrib.ox.ac.uk/fsl). The co-registeration was performed using the images with higher signal intensity and lower contrast (2 nd contrast from the Sa2RAGE and MP2RAGE sequences) and the spatial transforms were subsequently applied to the synthetic images. Re-sampling was performed using a sinc interpolation.
In the original MP2RAGE T 1 map calculation [5], look up tables were used to, assuming the transmit field was equal throughout the whole image, relate the MP2RAGE intensity to a T 1 value. A similar process was used to calculate B 1 + maps using the Sa2RAGE sequence [27], where an average T 1 value was assumed to be valid throughout the brain and was demonstrated to have a reduced impact in terms of the robustness of the calculated B 1 + maps. In this work, 2D lookup tables containing the T 1 values associated to certain MP2RAGE signal and B 1 + value (see Fig. 3a for the MP2RAGE protocol A) and the B 1 values associated to certain Sa2RAGE signal and T 1 value (see Fig. 3b referring to the Sa2RAGE protocol) were computed. A two dimensional interpolation was iteratively performed for each pixel using the two lookup tables. Given the higher independence of B 1 + estimation on the T 1 values (see Fig. 3b), the B 1 + was first calculated for each pixel assuming constant T 1 throughout the brain (1.5 s). These B 1 + values were then used to estimate the T 1 values via a 2D interpolation of the MP2RAGE lookup table (Fig. 3a). The process was repeated using the newly updated T 1 estimates for each voxel. At the third iteration, the variations in both B 1 + and T 1 were found to be under 10 23 .
To evaluate the effectiveness of the B 1 correction, the corrected R 1 maps (1/T 1 ) acquired with the high-resolution (and high CNR) protocol (ii-A) were compared with the low resolution (and low B1sensitivity) protocol (ii-B) in terms of general similarity between the distribution of R 1 values throughout white and grey matter.
The quality of the anatomical information present in high resolution R 1 maps was evaluated by observing the contrast and anatomical detail observed in structures such as the hippocampus. The individual R 1 maps were processed with freesurfer (http:// surfer.nmr.mgh.harvard.edu/) to create cortical surface models [31]. Five equally spaced surfaces were generated in between the white matter and pial surfaces in order to study cyto-architectonic cortical variations in the mid layer throughout the brain. A 2 mm smoothing along the surface was applied and the R 1 surface maps were then plotted using MATLAB (The MathWorks Inc.).

Results
Simulations showed that the parameters that optimise T 1 contrast between WM and GM are obtained by reducing the spacing between the two different inversion times (see experimental Protocol i-b). Figure 2a shows the lookup tables of the MP2RAGE signal intensity as a function of the T 1 values for the protocols with TR = 6 secs optimized for full T 1 range contrast or  WM-GM contrast (see Fig. 2a). Using partial k-space sampling in the slice encoding direction it was possible to reduce the number of excitations per GRE block and the sensitivity of the resulting image to transmit B 1 field inhomogeneities (note that in Protocol A a1/a2 = 4/5 while Protocol B a1/a2 = 7/7) (see dashed lines in Fig. 2a). Figure 2b shows the computed CNR between two tissues of successive T 1 values (spacing of 0.05 secs), as a function of T 1 for the different protocols. The CNR of the new sequence parameters for the range of T 1 values for which the optimization was performed (corresponds to the integral of the curves in Fig. 2b from 1.1 to 1.9 s) increased by 51%, which when taking into account the reduced number of excitations due to partial Fourier (PF) sampling CNR~CNR ffiffiffiffiffiffiffi PF p À Á is of ,33% (see Fig. 2b). Figure 4 shows midbrain MP2RAGE and MP2RAGE WMGM images. It is possible to see an increased delineation of the thalamus and its medio dorsal, ventral lateral and pulvinar nuclei (white arrows) as well as increase contrast within the brain stem (yellow arrows). It should be noted that the reduced intensity (darker MP2RAGE intensity) observed in cortical grey matter and deep brain structures does not reflect a reduction in SNR in those regions but simply the fact that the two images (GRE TI1 and GRE TI2 ) have similar amplitudes but opposite phases (see Fig. 2a). Table 1 shows contrast to noise values (calculated as in Eq. 2) between different regions of interest: Medial Thalamus vs outer thalamus; outer thalamus vs WM; Caudate vs the head of the Putamen; Brain Stem vs cerebellum white matter; on average the increase of the contrast to noise ratio increased by 34%, in good agreement with what was expected from simulations. Naturally, the MIP and Full Range image have contrast to noise properties that, from a quantitative perspective, are equivalent to those of MP2RAGE WMGM , (see Table 1). The main differences are the easier interpretability of the full range image (4 th column of Fig. 4) due to its conventional appearance (dark CSF and bright WM), while the MIP image (3 rd column of Fig. 4), having a similar appearance to Double Inversion recovery images naturally allow a tighter image display dynamic range. Figure 5 shows lookup tables associated with different protocols (ii.a and ii.b, see Methods section for more details) for 3 different relative B 1 + intensities (0.6, 1 and 1.4 times the nominal B 1 ). The increased B 1 + sensitivity of the protocol in Fig. 5a (ii.a) is related to the increase of the number of rf pulses used per GRE block and of their amplitude (one direct and one indirect consequence of increasing the spatial resolution of the T 1 maps and wanting to achieve optimum contrast). The protocol shown as insensitive to B 1 + has errors on the T 1 estimation that are always lower than 2.5% of the nominal value (for the range of T 1 values expected to be found in the brain at 7T) even in the presence of 640% deviations from the nominal B 1 + (see Fig. 5d). Such low sensitivity to the B 1 + inhomogenity (before any B 1 + correction) makes it a valuable protocol to validate the results achieved with the B 1 + correction procedure. Figure 6 shows different slices of R 1 maps of one subject (whose range 0.75-0.95 s 21 was adapted to enhance the variations observed within white matter) obtained with two different imaging protocols and at two different head and dielectric pads [28,29] positions (in order to create different B 1 interference patterns). The B 1 + maps, spatially co-registered to the R1 maps, are also shown (Fig. 6d,h) in a short range of relative B 1 values (0.3-1.4). It is possible to see that the major changes between the two B 1 maps are located in the parietal and cerebellum regions. Red arrows show regions where the spatial inhomogeneities observed in the uncorrected R 1 maps (Fig. 6b,c) were spatially correlated with low B 1 regions (in the parietal area) or high relative B 1 (increased contrast observed in the splenium corpus callosum). After correction, the high resolution R 1 maps (Fig. 6f,g) have increased similarity to the low resolution R 1 maps (Fig. 6a,e), with a more homogeneous and symmetric distribution of R 1 values throughout the brain. Regions where the B 1 was too low to achieve the adiabatic condition in the bottom of the cerebellum (white arrow) cannot be corrected with the proposed methodology. Note that the low resolution R 1 maps (Protocol ii-B) remain mostly unchanged before (Fig. 6a) and after correction (Fig. 6e), supporting its insensitivity to B 1 inhomogeneities. Highly myelinated white matter fibber bundles such the optic radiation and the Genu corpus callosum are enhanced with respect to the remaining white matter (blue arrows) even after correction. Figure 7 shows three different slices of R 1 maps of the human brain (the range, 0.45-0.80 s 21 , was set to enhance the variations observed within grey matter). In any of the R 1 maps it is possible to observe the expected distribution of R 1 in the cortex with higher R 1 values being present on primary sensory areas (singled out by blue arrows), such as the sensory motor, auditory and visual cortex, with respect to other neighboring regions. Nevertheless, it is also possible to observe in Fig. 7b (uncorrected R 1 maps) that many cortical inhomogeneities/asymmetries are observed (as pointed out by the red arrows) that are likely to be associated with the B 1 inhomogeneity and are indeed removed when taking the B 1 measurements into account (Fig. 7c). Figure 7d shows the inflated right hemisphere of an high resolution corrected R 1 map of one single subject across the middle layer of the cortex. Blue arrows show the same cortical regions as those shown on Figure 7c, with the primary sensory areas clearly having increased R 1 values when compared to the remaining cortex. Figure 8 shows an enlarged view of R 1 maps in the hippocampal region demonstrating that the high resolution of the obtained R 1 maps offers the possibility to use this quantitative information to visualize (the shape and separation between grey and white matter regions) and characterize its different fields and regions (with the dentate gyrus, DG, having the lowest R 1 values, followed by the CA1-CA3 and finally the CA4 and subiculum having higher R 1 values) [32]. The observation of the DG having a lower R 1 value than the CA fields is in good agreement with recent histology literature [33]. A clear delineation of all these structures is only discernible in the high resolution dataset and its relaxometry properties could show changes preceding the changes in volume often observed in pathologies.

Discussion
In this paper we have shown the ability to adapt the conventional T 1 weighted protocol used for whole brain imaging at 7T to focus on the tissues whose T 1 's lie in-between those of white and grey matter. Such adaptation allowed the visualization of thalamic nuclei and brain stem structures which were not visible in the conventional protocol because of lack of contrast to noise ratio (note that the observed contrast is still only related to the T 1 of the different tissues). Particularly, the increased sensitivity to changes of grey and white matter T 1 values could help detect or/ and characterize early stages of idiophatic Parkinson disease [34] that have been associated with deposition of Lewy bodies in the medulla oblongata and structural changes that progress in a caudorostral pattern [35,36], only affecting the substantia nigra in a stage where the first motor symptoms occur. Furthermore, the ability to recalculate a conventional T 1 -w contrast (where the contrast of subthalamic nuclei is still increased) could be used to obtain automatic segmentation of the whole brain, the thalamus and subsequently of its nuclei using available histology based atlas [37]. The better visualization or automatic delineation of some of Figure 6. Transverse and coronal slices of: (a) Uncorrected R 1 maps using protocol ii-B; (b,c) Uncorrected R 1 maps using using protocol ii-A with two different head and dielectric pad positions; (e) Corrected R 1 maps using protocol ii-B; (f,g) Corrected R 1 maps using using protocol ii-A with two different head and dielectric pad positions; (e,h) B 1 + maps corresponding to the two different head and dielectric pad positions; R 1 maps are shown in a short range from 0.75 to 0.95 s 21 to emphasize the sensitivity of the R 1 maps to the B 1 in-homogeneity, and the success of its correction. Arrows point out regions of increased difference between in the B 1 + maps that have clear implications on the uncorrected high resolution R 1 maps. Blue arrows point out white matter fibber bundles with increased R 1 values with respect to the remaining white matter. Red arrows point out regions of significant R 1 differences from the ground truth prior to the B 1 + correction, which are successfully corrected. The white arrow points out a region of very low B 1 + field where the adiabatic condition was not reached and hence the R 1 values are largely overestimated and not possible to correct with the proposed methodology. doi:10.1371/journal.pone.0069294.g006 these structures would allow a better localization of structures such as the such as the vim [6] which is of great importance in presurgery planning in pathologies such as in Parkinson's disease or dystonia [8,9].
In this work, it was demonstrated that the residual sensitivity to B 1 inhomogeneities present in the R 1 maps obtained with the MP2RAGE sequence can be removed using information from a B 1 map (the choice of the B 1 mapping technique is not crucial, and many other methods would be equally valid). In its current implementation, the B 1 correction to the R 1 maps was only considered to affect the excitation pulses and not the adiabatic inversion pulse, this could be introduced into the Bloch equation simulations by changing the inversion efficiency of the inversion pulse as a function of B 1 and B 0 map. This would be expected to have only a small effect on the R 1 values measured as one of the main characteristics of the adiabatic pulse used is its low sensitivity to B 1 and B 0 variations once the adiabatic condition is achieved [38]. The success of the correction presented in this manuscript allows to obtain R 1 maps in either: a reduced acquisition time (decreasing TR will now only have implication in terms of available contrast to noise [5]); increase contrast to noise ratio thanks to the possibility of using protocols with increased flip angles; increased resolution thanks to the possibility of using longer GRE blocks that would otherwise affect the B 1 insensitivity of the method.
High spatial resolution R 1 mapping at 3T has been shown to allow the identification of various cortical regions which have been validated by functional retinotopic and tonotopic studies [20,39]. The presented protocol to obtain quantitative R 1 values could be used in similar studies as there are remarkable similarities between the surface maps obtained (see Fig. 7d) with those found in literature, in which cortical regions were carefully validated with functional studies and the robustness of the methods were tested via test-retest of these regions in group studies. Although a comparison between the two methodologies is outside the scope of this paper (as the differences could result from the increased SNR, or R 1 dispersion available at 7T, or from the different efficiency of the R1 mapping methodologies), it should be noted that the present study was acquired with a higher spatial resolution (0.65 mm vs 0.8 mm isotropic), was obtained in a shorter amount of time (12 mins vs 21 mins) and the through layer smoothing of the surface map shown on Fig. 7d was significantly smaller (2 mm vs 4 mm) than the cortical R 1 maps shown in previous studies [20,39]. Two considerations can be made from this observation: the proposed protocol has the potential to obtain a delineation of the primary sensory regions on individual subject data as shown by Figure 7. Transverse, sagittal and coronal slices of: (a) a corrected R 1 map using protocol B and (b,c) uncorrected and corrected R 1 maps calculated using protocol A. The R 1 maps are shown in a short range from 0.45 to 0.80 s 21 in order to emphasize the sensitivity of the cortical R 1 maps to the B 1 in-homogeneity. Panel (d) shows the reconstructed corrected R 1 surface across the middle layer of the cortex in the right hemisphere as calculated by freesurfer. Blue arrows point out primary sensory cortices (visual, auditory and sensory-motor cortex) which can be associated with increased R 1 values. Red arrows point out regions of significant R 1 differences from the ground truth prior to the B 1 + correction, which are successfully corrected: on the coronal slice a strong left right asymmetry between similar cortical regions, while on the sagittal slice a strong anterior posterior variation is highlighted. doi:10.1371/journal.pone.0069294.g007 other groups; having a spatial resolution of the order of magnitude of the thickness of the different cortical layers (which have been demonstrated to have different relaxation properties) [7], makes the evaluation of R 1 maps at different cortical depths meaningful, such extra information could help further differentiating cortical regions [7,40] and will be the object of future research.

Future Work and Conclusions
We have shown that, by optimizing the contrast of the MP2RAGE sequence to a specific range of T 1 's at 7T, it is possible to gain access to clearer anatomical delineation of thalamic nuclei and brain stem structures. Such improved contrast could be an important asset in the context of automatic segmentation and in pre-surgical planning of Parkinson patients. One of the unanswered questions that will be the object of future work is the relationship between the observed sub-structures of the thalamus observed in T 1 -w imaging and the sub-thalamic nucleus identified with histology so that the automatic segmentation of the thalamus can be guided by the image contrast rather than simply the strength of the given prior.
We demonstrated the ability to obtain high resolution (0.65 mm isotropic) and high SNR R 1 maps of the whole brain in ,12 mins using the information from separately acquired R 1 and B 1 maps from the MP2RAGE and Sa2RAGE sequences. The quantitative R 1 maps revealed subtle differences between distal GM tissues as shown in previous studies, but also within WM where R 1 contrast is usually overlooked. and coronal (c) slices covering the hippocampus of a high resolution corrected R 1 map. Arrows show hippocampal structures: CA1, CA2 3 (blue arrow); fimbria of hippocampus (yellow arrow); CA4 and DG (black arrow); subiculum (green arrow) [32]. Such structures are only discernible in the high resolution dataset. doi:10.1371/journal.pone.0069294.g008