- 1 The author is grateful to Dr. Thierry Delzescaux for the collaboration work done on 3D histology, (...)
1The 3D histological imaging of primate brain is the result of the stacking and alignment of serial histological slices spanning the brain or a subregion of it so that to recreate a consistent 3D volume.
2The creation of the 3D histological volume is a multi-steps process. After the sacrifice of the animal, the brain is extracted and brain tissues are undergoing histological preparation (such as paraformaldehyde fixation, paraffin embedding). They are then sliced using a cutting device such as a cryostat or a sliding microtome, the resulting sections can be spread on glass slides, stained, then they are finally digitized using an acquisition device (flatbed scanner, camera) to generate electronic image files, similar to the individual files corresponding to slices for a Magnetic Resonance Imaging (MRI) scan for instance.
3These images are then stacked and aligned so that each slice is best superimposed onto the following slice according to a certain similarity metric (possibly involving an external reference) and using a certain category of transformations.
4By propagation, the resulting volume thus becomes 3D spatially consistent, which implies in particular that it can be observed in any orientation. A normalization of the intensities within or between slices can be performed to correct for potential image intensities inhomogeneities.
5Depending on what needs to be done with the resulting volume, one can just use this volume as it is and perform for instance region of interest (ROI) based studies comparing a 3D region from one hemisphere to the corresponding one in the other hemisphere, just like what is classically done on intrinsically 3D medical images. Alternatively, one can aim at correlating the histological volume to an in vivo medical device image such as an MRI scan. In that case, a warping transformation has to be estimated to correct for the deformations due to the histological processing undergone by the brain (including severe shrinkage, see Figure 1) so that to match any point in the histological volume to the corresponding point in the MRI.
A stained histological slice of a baboon brain (a), the corresponding slice on the MRI (b) and a fusion image with the histological slice in false colors orange superimposed on the MRI at the same scale (c): the histological preparation induced a severe shrinkage to the brain compared to its in vivo geometry quantitatively evaluated at 15% in this case (images and data from Dauguet, 2005).
Une coupe histologique marquée de cerveau de babouin (a), la coupe correspondante de l'examen d'Imagerie par Résonance Magnétique (IRM) (b) et une image de fusion représentant la coupe histologique en fausses couleurs oranges superposée sur l'IRM à la même échelle (c) : la préparation histologique a induit une contraction importante du cerveau par rapport à sa géométrie in vivo évaluée à 15% dans les données de cette étude (d'après Dauguet, 2005).
6Thanks to the disease models widely developed and the similarity of their brain with humans (Figure 2), primates are a privileged specie to perform this 3D histological reconstruction and in vivo - post mortem correlation to take full advantage of the biological data provided by each animal involved in a study.
Brain MRI images in axial view of a macaque (a), a baboon (b) and a human (c) located approximately in the same anatomical level. The brains show many similarities at this level of observation which suggest why non-humans primates are widely used for pre-clinical studies (images and caption from Dauguet, 2005).
Examens IRM cérébraux en incidence axiale de macaque (a), babouin (b) et humain (c) observés approximativement au même niveau anatomique. Les cerveaux apparaissent relativement similaires à ce niveau d'observation ce qui suggère pourquoi les primates non-humains sont fréquemment utilisés dans les études pré-cliniques (d'après Dauguet, 2005).
7Different methods have been proposed to achieve this correction, some of them being simple and performing a global matching whereas some others are very complex and can even perform one to one voxel correlation.
8In a first part, we will describe in details the different steps of the 3D histological imaging and the solutions that have been proposed in the literature, we will then give an overview of the applications using the 3D histological imaging in the second part, before considering some possible future developments in the third part of the article.
9Generally speaking, for primate brains, it can be considered that there are two main types of brain deformations caused by histological preparation and fixation: global 3D deformation and slice-specific 2D deformations, following the classification proposed by Dauguet (2005).
10Global 3D deformations caused by the extraction of the brain from the skull include (1) global shrinkage due to the loss of cerebrospinal fluid, loss of blood irrigation, and global dehydration, and (2) various deformations due to gravity or other mechanical effects, and possible specific processings (such as fixation and paraffin embedding). We defined these global three-dimensional deformations as primary deformations because they are first to appear and they occur before sectioning the brain. In order to correct these deformations, it is necessary to estimate a 3D transformation between the brain in its post mortem geometry and the actual in vivo brain.
11The individual 2D deformations for each slice are due to the serial sectioning of the brain into 2D slices, and include: (1) specific deformations due to cutting (shearing, tearing), (2) shrinkage due to temperature changes (see Gardella et al., 2003, for a quantitative study of deformations), and (3) changes in the original geometry of each slice due to the mounting on glass slides of both hemispheres and displacements of smaller parts like gyri. We defined these individual 2D deformations as secondary deformations because they appear during the cutting step, after the primary deformations. The correction of these deformations requires first the estimation of the initial spatial 2D configuration of each slice, second the reconstruction of the 3D volume by alignment of the series of slices.
12Figure 3 illustrates some of the steps of the histological preparation and the corresponding deformations.
Examples of the deformations undergone by the primate brain during histological preparation that have to be corrected to match the series of histological slices with an in vivo device image. The brain is extracted from the skull (a), it can be divided in smaller blocks to favor fixation (b): these deformations are 3D deformations, or primary deformations. The brain is then sliced using a cutting device like a sliding microtome (c), then sections are stored in small containers (d) before being mounted on glass slides and stained (e): the latter deformations are 2D deformations, or secondary deformations(images and caption from Dauguet, 2005).
Exemples de déformations subies par le cerveau de primate pendant la préparation histologique qui doivent être corrigées pour mettre en correspondance la série de coupes histologiques avec un examen d'imagerie in vivo. Le cerveau est extrait de la boîte cranienne (a), il peut être divisé en blocs pour améliorer la fixation (b) : ces traitements génèrent des déformations 3D dites déformations primaires. Le cerveau est ensuite coupé avec un appareil de coupe tel qu'un microtome à glissière (c), puis les sections sont stockées dans des puits (d) avant d'être montées sur lames de verre et marquées (e): ces derniers traitements génèrent des déformations 2D dites déformations secondaires (d'après Dauguet, 2005).
13Since the creation of the 3D histological imaging starts with the series of histological slices as input material, the correction of the deformations needs to be performed backward. Secondary deformations are corrected first to get a 3D volume, then primary deformations are corrected to correlate the histological slices with in vivo images.
14The correction of these two different types of deformations that will ultimately link the analysis of histology and MRI can benefit from the use of an intermediate modality. This intermediate modality consists of photographs or videos of the brain surface taken during the sectioning process (Toga et al., 1994; Kim et al., 1997; Mega et al., 1997; Ourselin et al., 2001; Dauguet et al., 2007; Bardinet et al., 2008), which are usually called blockface photographs (Figure 4).
Blockface photographs of a baboon brain (a) and of a macaque brain (b) during cutting with a sliding microtome (images from Dauguet, 2005).
Photographies de la surface d'un bloc cerveau de babouin (a) et de macaque (b) en cours de coupe au microtome à glissière (d'après Dauguet, 2005).
15This intermediate modality has 2 main properties, due to the fact that each photograph of a slice is taken prior to sectioning, which are essential for the correction of both primary and secondary deformations.
16First, the imaged section on the photograph does not present (or minimally) the 2D secondary deformations such as shearing due to cutting, independent movements of individual parts of the brain (hemispheres, cerebellum etc...) or other mechanical distortions due to histological procedures (mounting on glass slides for instance).
17Second, if the photographs are taken repetitively with the brain in the exact same position, a simple stacking of the photographs provides a direct 3D reconstruction of the series of photographs of the brain with no need for slice-to-slice registration. The reconstructed volume obtained this way has the advantage of preserving the natural 3D curvature of the brain which tends to be lost when an a posteriori alignment using slice-to-slice registration is performed. This « banana » effect as described in Streicher et al. (1997) and Malandain et al. (2004) is illustrated in Figure 5.
The banana problem: the 3D reconstruction of a 3D curved object is not easy. (a) Take a 3D curved object (e.g., a banana) (a); cut it into slices (b); digitize the slices (c); mix the digitized slices (d); the 3D reconstruction results in a cylindrical banana (e). Using a shape prior (e.g., MRI) (f) may help to reconstruct the curved banana. (images and caption from Malandain et al., 2004).
Le paradoxe de la banane : la reconstruction 3D d'un object 3D incurvé n'est pas tâche facile. Considérons un tel objet (une banane par exemple) (a), coupons la en tranches (b), digitalisons les coupes (c), décalons arbitrairement les coupes digitalisées (d) : la reconstruction 3D obtenue est une banane de forme cylindrique (e). L'utilisation d'un a priori de forme (par exemple un examen IRM) (f) peut aider à reconstruire la banane avec sa courbure originale (d'après Malandain et al., 2004).
18The photographic volume when available is thus often used as a reference for the alignment of the series of histological slices, and for the estimation of the warping transformation to correct for the deformation of the post mortem brain.
19In addition to serving as an intermediary modality, this volume has other advantages by itself. It can be used directly as a source of information for a biomedical study. This is particularly useful when the brain was stained prior to cutting so that the photographs show the staining (see Figure 6 from Annese et al., 2009), or when only information of the precise location of an injection (needle) or a point of measurement (electrode) is needed, since the track of the device is clearly visible in 3D (see Figure 7 from Drouot et al., 2004).
A macaque brain which was perfused in situwith methylene blue and paraformaldehyde fixative following a special procedure. Photographs of the blockface taken during sectioning of a macaque brain. We can see the pink freezing medium (a), the same image after automatic segmentation of the slice (b) and volume reconstruction showing 3 orthogonal planes of section crossing at the level of the basal ganglia. The insert shows a detail of the sagittal plane across the cerebellum that has been resampled from the original volume created with coronal sections (c) (images and caption from Annese et al., 2006).
Cerveau de macaque perfusé in situ avec du bleu de méthylène et une solution de paraformaldéhyde selon une procédure spéciale. Photographies du cerveau prises au cours de la coupe. On peut voir le milieu congelant de couleur rose enrobant le cerveau (a), la même image après segmentation automatique de la coupe (b) et la reconstruction volumique mettant en évidence 3 plans orthogonaux de coupes se croisant au niveau des noyaux gris centraux. L'encart montre un agrandissement d'un plan sagittal traversant le cervelet qui a été rééchantillonné à partir du volume original constitué de coupes coronales (c).
Slices from the photographic volume of a baboon brain obtained after 3D reconstruction of the series of photographs. The volume has been resampled to show the entire course of the electrodes which have been inserted in the internal globus pallidus (a) and the subthalamic nucleus (b) to measure neurons activity (images and caption from Drouot et al., 2004).
Coupes extraites du volume photographique d'un cerveau de babouin obtenues après reconstruction 3D de la série de photographies du bloc cerveau en cours de coupe. Le volume a été rééchantillonné pour mettre en évidence la descente des électrodes qui ont été introduites dans le globus pallidus interne (a) et dans le noyau sous-thalamique (b) pour mesurer l'activité des neurones (d'après Drouot et al., 2004).
20The correction of the secondary deformations previously described leads to the 3D reconstruction. Many works have been done on the alignment of series of 2D slices.
21The first category of works uses global linear transformations such as rigid or affine transformations (Hibbard and Hawkins, 1988; Andreasen et al., 1992; Zhao et al., 1993; Goldszal et al., 1995; Schormann et al., 1995; Rangarajan et al., 1997; Cohen et al., 1998; Ourselin et al., 2001; Malandain et al., 2004; Dauguet et al., 2005; Bardinet et al., 2008). These global transformations potentially include translations, rotations, scale factors and shearing. They are easy to compute, robust and are usually invertible. On the other hand, the correction is global and local deformations between successive slices will not be corrected by this class of transformation.
22Non-linear transformations can be used for greater flexibility so that to correct more subtle inter-slice differences. Non linear transformations can be divided into parametric transformations, which are transformations that can be represented by a finite number of parameters (like cubic B-splines, see Rueckert et al., 1999 for instance), and non-parametric transformations for which a displacement vector needs to be estimated for each pixel of the image independently (like the demons algorithm, see Thirion et al., 1998 for instance). While these transformations can provide visually very smooth 3D reconstructions, and achieve a high degree of inter-slice correlation, the iterative use of non linear-transformation on long series of slices can become problematic in terms of robustness and stability. They are also more sensitive to the kind of artifacts that are very often observed on histological material (holes, missing parts, torn tissue, foldings). Some works have successfully applied these non-linear transformations to correct secondary deformations for the alignement of histological slices (Kim et al., 1997; Mega et al., 1997; Schormann and Zilles, 1998; Chakravarty et al., 2006; Ceritoglu et al., 2010).
23Contrary to rodents, a particularity of most primate brains - if we except small lemur primates like Microcebus murinus - is that the brain is circumvoluted, which means its surface is made of sulci (fissures) and gyri (lobes) (Figure 8).
Differences of the degree of circumvolution of the surface of the brain. Almost flat for a rodent (a), deep sulci for a primate (b). The deformation of the slice during the histological preparation is therefore different (images from Dauguet, 2005).
Différents degrés de circonvolution de la surface corticale du cerveau : presque plat pour un rongeur (a), présentant de profonds sillons pour un primate (b). Les déformations engendrées par la préparation histologique sont par conséquent différentes pour ces coupes.
24For this reason, some whole parts of the brain, like a girus or even a whole hemisphere, can be somehow articulated and have independent displacements during the histological preparation, the cutting and the mounting on glass slides (Figure 9).
The photograph blockface image of a slice before cutting (a), the same slice after cutting and mounting on a glass slide (b) and the superimposition of the mounted slice in false color orange onto the photograph (c): we can see clearly the independent displacement of the hemispheres on the mounted slice. Another example of a mounted histological slice (false color orange) superimposed on the corresponding photograph (d): this time, the gyrus in the white frame clearly moved independently of the rest of the slice (images and caption from Dauguet, 2005).
Photographie de la surface du bloc cerveau d'un primate prise avant le passage du couteau (a), coupe histologique extraite, correspondant à la surface photographiée, montée sur lame et marquée (b) et superposition de la coupe histologique en fausses couleurs oranges sur la photographie (c) : on peut clairement noter le mouvement indépendant des hémisphères de la coupe histologique par rapport à la photographie. Un autre exemple de coupe histologique montée sur lame (fausses couleurs oranges) superposée sur la photographie correspondante de la surface du bloc cerveau (d) : cette fois, on peut voir que le gyrus dans le cadre blanc a bougé indépendamment du reste de la coupe (d'après Dauguet, 2005).
25Some works tackle these problems specifically by estimating local linear transformations for specific independent moving parts (Arsigny et al., 2005; Pitiot et al., 2006; Dauguet et al., 2007). The blockface volume can be used as a reference for the positioning of the parts within each histological slice. These classes of transformations have the advantage of being both more flexible than classical linear transformations and more controllable than pure elastic transformations. They are thus well adapted to the correction of the secondary deformations for histological slices of primates.
26The correction of the primary deformations leads to the 3D post mortem - in vivo matching. To estimate 3D transformations for the registration of two brains, several techniques have been proposed in the literature, either performing rigid transformations using mutual information as similarity criterion (Viola and Wells, 1997; Maes et al., 1997), or non-linear transformations (Ashburner and Friston, 1997; Friston et al., 1995; Davatzikos, 1997; Collins and Evans, 1998; Thirion, 1998; Christensen, 1999; Cachier et al., 2003; Clatz et al., 2005). More complete overviews of medical image registration can be found in Maintz and Viergever (1998) and Salvi et al. (2007). For the particular 3D registration of post mortem and in vivo primate brains, different strategies have been proposed in the literature. Ourselin et al. (2001) and Bardinet et al. (2008) have proposed an affine 3D transformation for the registration of the basal ganglia. For global matching of one hemisphere, Malandain et al. (2004) proposed a method based on an alternated correction of primary and secondary deformations using 3D and 2D affine registrations. Schormann et al. (1995) used an optimized affine registration of labelled regions to correct for the deformations between histological slices and the MRI section. Thompson and Toga (1996) developed a 3D warping method based on surface that was also used in 2D for histology registration in Mega et al. (1997). A 3D viscous fluid transformation model has been proposed by Christensen et al. (1997) and applied to the deformation of blockface photographs, whereas Schormann and Zilles (1998) performed a 3D transformation based on the elastic medium theory with a full-multigrid strategy. Elastic methods using thin plate splines with control points have been proposed to be used for rodent brains (Kim et al., 1997 ), which do not, however, present the same deformations as that of primate brains. Dauguet et al. (2007) proposed to use cubic B-spline transformations with mutual information to match in vivo and post mortem images. Recently, Ceritoglu et al. (2010) proposed to use of large deformation diffeomorphic mapping registration for the correction of the primary deformations. In their case, they did not have acquired blockface photographs.
27The 3D histological imaging of primate brains, thanks to its geometrical properties and its 3D consistency, allows one to perform two main applications that can not be realized on subset or series of histological slices with no 3D reconstruction.
28First of all, it extends the possibilities of studies of histology thanks to the possible use of 3D techniques of analysis up to yet only used on intrinsically 3D medical images. We can cite for instance the delineation of 3D regions of interest respecting natural 3D symmetry of the brain for objective measurements, or the voxel-wise statistical analysis which allows exploratory research with no biological a priori on a particular region involved in the process studied.
29Second, it makes possible the matching with classical in vivo medical device images, used in particular for humans in clinical routine. The direct consequence of this is that it becomes possible to compare the signal of the imaging devices to the signal in the same location on the histology, which thanks to the many available stainings, is often considered as the gold standard in terms of information.
30This first application takes advantage of the spatial consistency of the 3D histological imaging to create atlases with both histological precision and full 3D consistency. The 3D consistency of an atlas enhances its interest compared to a paper atlas (like the reference Paxinos et al., 2000) since it becomes possible to reslice the atlas in any desired plan suitable for the observation of a structure of interest. Furthermore, in case an anatomical 3D image (e.g. an T1 MRI or a 3D histological imaging) is available for a given primate involved in a study, it is possible to map the general digital 3D atlas so that it fits the structure features of this particular primate.
31In the article of Dauguet et al. (2009), the author describes a framework to generate a 3D digital atlas of the thalamus based on a series of stained histological sections of a primate. The contours of the thalamus were first drawn on the stained histological slices. The series of histological sections were then aligned and mapped onto the in vivo MRI of the same animal acquired prior to the sacrifice using the blockface photographs as an intermediary modality. By applying the series of transformations previously estimated for the histological volume, the contours of the digital atlas were mapped onto the MRI data (Figure 10).
Drawn contours of the thalamus atlas of a baboon brain superimposed on the projection of the stained histological section serving as model as it is seen through the camera lucida (a), the final paper atlas drawing of the considered section (b), a surface rendering of the whole digital atlas based on the drawn sections (c), the 3D histological volume which served as model for the atlas (in false colors red) correlated and sumperimposed on the corresponding MRI (d) and the surface rendering of the digital thalamus atlas mapped onto the MRI (e) (images and caption from Dauguet et al., 2009).
Tracés des contours de l'atlas du thalamus d'un cerveau de babouin superposés à la projection de la coupe histologique servant de modèle tels qu'observés par camera lucida (a), les contours de l'atlas papier de la coupe considérée obtenus (b), rendu surfacique de l'atlas digital entier obtenus à partir des tracés papier (c), le volume histologique 3D qui a servi de modèle pour la réalisation de l'atlas (en fausses couleurs rouges) mis en correspondance et représenté superposé à l'IRM correspondante (d) et le rendu surfacique de l'atlas digital du thalamus repositionné dans l'espace de l'IRM (e) (d'après Dauguet et al., 2009).
32The protocol was tested on baboon brains for which the full series of slices were available, but also on a macaque brain for which a subset only of the histological slices were available demonstrating the ability of creating hybrid 3D histological imaging (see Figure 11) and building digital atlases in the MRI geometry without mounting and staining all the brain slices. The accuracy of mapping the digital atlas of one baboon onto the MRI of the other baboon by comparing the overlapping with its original digital atlas was then studied and demonstrated that the warping of a general 3D digital atlas onto a specific individual can be performed with good accuracy. The digital atlas of one of the baboons was finally used to study the individual kinetic of the main thalamus nuclei on Positron Emission Tomography (PET) images providing a novel and accurate way of measuring very fine and local functional differences.
Hybrid 3D reconstruction made of both photographs and histological slices (1st column); corresponding MRI image (2nd column); fusion of the hybrid volume (false colors orange and blue) and the MRI image (3rd column) of a macaque brain in coronal (1), axial (2) and sagittal (3) views. We can see that the subset of stained histological slices could be corrected from the deformations they undergone during histological preparation and correlated with the MRI thanks to the the photographs which completed the volume (images and caption from Dauguet, 2005).
Reconstruction 3D hybride constituée à la fois de photographies et de coupes histologiques (1ère colonne); examen IRM correspondant (2e colonne); fusion du volume hybride (fausses couleurs oranges et bleues) et de l'examen IRM (3e colonne) d'un cerveau de macaque en incidence coronale (1), axiale (2) et sagittale (3). On peut voir que le sous-ensemble de coupes histologiques marquées a pu être corrigé des déformations subies pendant la préparation histologique et a pu être mis en correspondance avec l'IRM grâce aux photographies qui ont complété le volume (d'après Dauguet, 2005).
33Such atlases are very promising and provide efficient ways of performing segmentation of the brain with high precision. We can also cite similar applications of the histological imaging based atlas for humans (Chakravarty et al., 2006; Yelnik et al., 2008) among others, or a slightly different use of the histological imaging for rodents - potentially feasible on primates - where this time an MRI-based atlas is used on histological data (Lebenberg et al., 2010).
34This second application of the 3D histological imaging takes advantage of the possibility of correlating the in vivo MRI of a primate to the corresponding histological slices which will serve as a gold standard. In this case, the technique studied for validation is the Diffusion Tensor Imaging (DTI) based white matter tractography. Since the introduction of diffusion weighted imaging (DWI) as a method for examining neural connectivity, its accuracy has not been formally evaluated. In Dauguet et al. (2007a), the authors directly compared connections that were visualized using injected neural tract tracers (wheat germ agglutinin conjugated to horseradish peroxidase: WGA-HRP) with the connections obtained using in vivo diffusion tensor imaging (DTI) tractography. First, the tracer was injected at multiple sites in the brain of a macaque monkey; second, the histological sections of the labelled fiber tracts were reconstructed in 3D; third, the fibers (somatosensory and motor tracts) were segmented and registered with the anatomical in vivo MRI from the same animal; and last, fiber tracing was conducted along the same pathways on the DTI data using a classical diffusion tracing technique with the injection sites as seeds. To evaluate the performance of DTI fiber tracing, the fibers derived from the DTI tractography were compared to those segmented from the histology. The influence of the parameters controlling the tractography was also studied by comparing Dice superimposition coefficients (Dice, 1945) between histology and DTI segmentations. While there was generally good visual agreement between the two methods (see Figure 12), the quantitative comparisons revealed certain limitations of DTI tractography, particularly for regions at remote locations from seed, which thus demonstrated the importance of appropriate settings for realistic tractography results.
Comparison of the white matter tracks (descending axons from primary cortex through the internal capsule, ascending thalamocortical fibers to somatosensory cortex and transcallosal pathways) revealed by neural tracers (WGA-HRP) directly segmented on the 3D reconstructed histological imaging volume (in blue), with the results obtained from performing Diffusion Tensor based Tractography using a standard hyperstreamline-based method (in brown) within a half-transparent 3D rendering of the brain for a macaque in coronal (a), axial (b) and sagittal (c) orientations. Although the segmentations show some clear similarities, there are still some big differences including some extra pathways oversegmented by the DTI based method (images and caption from Dauguet et al., 2007a).
Comparaison de faisceaux de matières blanches (axones issus du cortex primaire traversant la capsule interne, fibres thalamocorticales se projetant dans le cortex somatosensoriel et connexions transcalleuses) révélés par traceur neuronal (WGA-HRP) et segmentés directement sur le volume histologique 3D reconstruit (en bleu), avec les résultats obtenus par tractographie basée sur l'imagerie du tenseur de diffusion en utilisant une méthode standard de suivi de direction principale (en marron) représentés à l'intérieur d'un rendu surfacique semi-transparent du cerveau d'un macaque en incidence coronale (a), axiale (b) et sagittale (c). Bien que les deux types de segmentation apparaissent assez semblables, il subsiste des différences importantes comme par exemple l'existence de certains faisceaux supplémentaires sursegmentés par la méthode de tractographie de diffusion (d'après Dauguet et al., 2007a).
35We described how to create a 3D spatially consistent volume from a series of histological slices of primates and how it could be correlated with in vivo medical device images such as an MRI. The 3D histological imaging bears 2 main properties. It is 3D spatially consistent and it can be made coherent with the brain in its in vivo geometry.
36Thanks to its 3D spatial consistency, the 3D histological imaging gives access to the processing and analysis tools classically dedicated to intrinsically 3D images such as Magnetic Reasonance Imaging (MRI), Positon Emission Tomography (PET) or Computer Tomography (CT) scans. We already talked about 3D elastic registration applied to the 3D histological volume; similarly the classical 3D segmentation methods can as well be applied to this volume (region growing, level set, atlas warping) to extract in 3D structures of interest. Another very promising research field is certainly the voxel wise statistical studies and voxel based morphometry as perfomed by software like the widely used Statistical Parametric Mapping (SPM). Some rare but promising studies have already been conducted on rodents (e.g. Dubois et al., 2008), and there is no reason why it could not be tested similarly for primate 3D histological volumes with great added value on objectivity and refinement of the analysis.
37Thanks to its coherence with the brain in its in vivo geometry, the 3D histological imaging gives access to direct comparison of the histology with the medical device images signal, and thus to the potential validation of the physical signal measured, by the direct optical observation of tissues. This is thus of high interest to validate new radiomarkers dedicated to the early detection of a pathology or reconstruction methods for PET scanner, or new sequences, new contrast agents for MRI devices for instance. In this latter category of validation experimentation, we can cite a potentially very informative comparison that could be performed on lemur primate (Microcebus murinus) model of aging process and Alzheimer disease between the special MRI sequences that produce promising results to detect amyloid plaques (Dhenain et al., 2000), with the corresponding histological slices processed with a staining (like Congo Red) which specifically marks the amyloid plaques.
38If we take the principle of the 3D histological imaging to the next level, we step into the microscopy level of observation. This means that instead of using a classical flatbed optical scanner or a CCD camera for the acquisition of the histological sections, each histological slice could be observed via a microscope offering high magnification and potentially very thin sections. Note that the observation can be done using optical microscopy, but also fluorescence microscopy or electron microscopy for instance. One of the many difficulties of this type of observation with high magnification is the size of the images produced, which implies a new way of processing them since the software classically used to process the biomedical data are not adapted to such data. On the other hand, the quantity of information that could be derived from the 3D histological volume made of microscopy scanned images would be tremendous and would allow to link microscopical physiological processes at the cell level to macroscopical measurements. Some tests have already been done on ferret on a small region with electron microscopy (Dauguet et al., 2007b) but as of today, the tools and technical capabilities to cope with such volumes are not available yet for primates.
39Finally, it is important to specify that this technique can be extended to body parts other than the brain (we can think of the heart for instance as it is being intensively studied) and species other than primates, so it is a rather general form of imaging similarly to MRI or CT scans.