Iron is an essential cofactor in many physiological processes. In the brain, it contributes to myelination, DNA synthesis, mitochondrial respiration, oxygen transport, and neurotransmitter metabolism (Ward et al., 2014). The iron redox couple Fe²⁺/Fe³⁺ mediates many reactions, the precise redox potential varying according to the nature of the ligand. Fe²⁺ is essentially present in the labile iron pool in the cytosol, which represents the iron that is directly available for cellular chemistry. Fe³⁺, which is insoluble, is the major form of iron in the brain, and is mainly found in iron-binding proteins. Hemoproteins (hemoglobin, myoglobin, cytochromes) represent a large proportion of iron in the whole body, but only a minor proportion of iron in the brain parenchyma, except in large vessels. Non-heme iron proteins can be classified according to their function. Transferrin (TF) is the main protein involved in iron uptake by the transferrin-transferrin receptor pathway. TF has two Fe³⁺ binding sites, but in physiological conditions, only 30% of TF molecules are loaded (Hare et al., 2013). Transferrin receptor (TfR 1 and 2) are expressed in all the central nervous system and especially in the plasma membrane of the neurons. The non-transferrin bound iron pathway is a minority pathway including several channel proteins and ligand/receptors interactions, notably Divalent Metal Transporter 1 (a channel protein) responsible for ferrous iron uptake. Ferroportin (FP) is the only known iron export protein. FP exports Fe²⁺ from the cytosol, and Fe²⁺ is further oxidized by hephaestin or ceruloplasmin and their ferroxidase activity. Both are expressed in the brain (Jiang et al., 2017). Hepcidin regulates systemic iron homeostasis by direct interaction with FP. In the cytosol, except in the labile iron pool, iron is ligated by low molecular weight proteins such citrate or ascorbate, iron-sulfur proteins, or ribonucleotides. Ferritin is the major storage protein. It is composed of two subunits, heavy (H) and light (L) chain, that surround a ferrihydrite core capable of storing up to 4,500 iron atoms. In physiological conditions, however, only 2,000–2,500 atoms are usually stored (Bossoni et al., 2017). It should be noted that ferritin may also be iron-free, especially in a pro-inflammatory condition. In the brain, oligodendrocytes (cells producing myelin) are the cells with the most intense immunostaining for ferritin in normal conditions because of their high iron metabolism for axonal myelination. They express both L and H subunits, like microglial cells. The neurons essentially express H subunits. The composition of ferritin in astrocytic cells has yet to be determined (Jiang et al., 2017). L-subunit is required for the long-term storage of iron, whereas H-subunit has ferroxidase activity. The H/L ratio increases with age by a factor 2–3 (Connor et al., 1995).

Magnetite was recently discovered in the brain, and may substitute the ferrihydrite crystal in ferritin in pathological conditions such as Alzheimer's disease (AD) (Bossoni et al., 2017), although this is quite controversial, as some authors claim that the magnetite crystal is too large (Kumar et al., 2016). Magnetite has a magnetic moment 10 times greater than that of ferrihydrite, but seems to be a thousand times less concentrated (Bulk et al., 2018b). Despite its rather misleading name, hemosiderin contains non-heme iron (Harrison and Arosio, 1996). Hemosiderin is regarded as a ferritin degradation product, as immunolabeling studies have found ferritin expressed in siderosomes. Siderosomes are structures that appear to be derived from lysosomes that have degraded ferritin to form hemosiderin in excess iron conditions. In the brain, hemosiderin can be also found during hemorrhages (hematomas, microbleeds, etc.) or in pathological conditions involving excess iron (Harrison and Arosio, 1996; Quintana et al., 2006). Iron is mainly present in the form of ferrihydrite crystals, as it is in ferritin. Hemosiderin is regarded as insoluble, unlike ferritin, and has a high iron/protein ratio. The protein part is less clearly structured than that of ferritin.

Neuromelanin and lipofuscin are two lysosomal products. Neuromelanin is a black, insoluble pigment incorporated into 0.5–3 μm intracytoplasmic organelles. These organelles are composed of proteins, lipids and neuromelanin granules of 200–400 nm with 30-nm substructures. Although the composition and appearance of neuromelanin vary in electron microscopy, depending on the brain region from which it is extracted, the 30-nm substructure is remarkably well-preserved (Zecca et al., 2008; Bush et al., 2009; Engelen et al., 2012). It comprises a eumelanin envelope with a low oxidative potential surrounding a pheomelanin core, with a pheomelanin/eumelanin ratio of 3/1 (Zecca et al., 2008). Neuromelanin is now known to be produced by catecholaminergic neurons in response to excessive cytoplasmic accumulation of unsupported catecholamines in synaptic vesicles. Aside from its role in catecholamines catabolism, neuromelanin is a chelator of heavy metals and environmental toxins (Zucca et al., 2017). In terms of absolute concentration, the most chelated metal by far (by a factor of at least 10³ relative to the others) is Fe³⁺. However, the number of chelated metals is extremely varied. It is in the melanin-containing neurons of the susbtantia nigra that we find the most neuromelanin, with a higher iron concentration than elsewhere. These neurons contain very little ferritin, compared with non-melanin-containing neurons and adjacent glia (Zecca et al., 2004). Neuromelanin has two iron binding sites: a high-affinity site for storing iron in an inactive redox form, which thus participates in its antioxidant action; but also a lower-affinity site that comes into play when the cell is in an iron overload situation. This site is responsible for the easier release of iron and thus contributes to the pro-oxidative properties of neuromelanin. It is now generally accepted that neuromelanin has a dual action (Faucheux et al., 2003; Zucca et al., 2017). There has recently been renewed interest in this substance, partly because of its involvement in the pathophysiology of Parkinson's disease (PD) and related pathologies, and partly because modern imaging techniques allow it to be detected and quantified in vivo, and can thus yield evidence of its variations as a function of time and pathology. This is the case in magnetic resonance imaging (MRI) with fast spin echo techniques involving magnetization transfer (Sasaki et al., 2006, 2008) as well as in positron emission tomography (PET) (Hansen et al., 2016). The iron concentration of neuromelanin has also been experimentally shown to be correlated with T1 and T2 relaxation (Trujillo et al., 2016).

Lipofuscin is another insoluble and ubiquitous pigment (yellow-brown) of the brain. Lipofuscin differs from neuromelanin in its autofluorescence with ultraviolet light (Double et al., 2008). It is also the result of autophagocytosis by the lysosomal system, including autophagocytosis of mitochondria. The final organelle therefore has a size comparable to that of neuromelanin. It is rather perinuclear compared with neuromelanin, which can be found at axonal endings. Lipofuscin increases with age and is a marker of neuronal aging. It is also involved in the chelation of metals such as iron. The presence of free iron contributes to an increase in lipofuscin, probably through lipid peroxidation (Ashraf et al., 2018). Although they would appear to be quite similar, in terms of their formation mechanism and their structure (proteins, lipids, metal ions) lipofuscin and neuromelanin actually seem quite distinct. Moreover, there is no lipofuscin in neurons containing neuromelanin (Terman and Brunk, 2004; Double et al., 2008; Sulzer et al., 2008). There are no scientific data on the potential role of lipofuscin in relaxation in MRI and there are no specific designed sequences for lipofuscin compare to neuromelanin, but we can assume that it does indeed play a role, by virtue of its nature and its proximity to neuromelanin.

In physiological conditions, iron homeostasis is strictly regulated, but in many pathological processes, it is dysregulated, leading to the formation of reactive oxygen species implicated in oxidative stress. The main pathologies concerned are extrapyramidal disorders such as PD or multiple system atrophy (MSA), AD, multiple sclerosis (MS), and specific iron metabolism disorders grouped under the heading of neurodegeneration with brain iron accumulation (Dusek et al., 2012). Aging is the main factor favoring accumulation of iron in the brain. Iron increases during the first decades, reaching maximum values during the fifth or sixth decade with only subtle changes after (HALLGREN and SOURANDER, 1958).

Ever since the pioneering work carried out around the turn of the Twentieth-century by Zaleski (1887) and Spatz (1922) (Koeppen, 1995), brain iron quantification and characterization has been an important topic of research. Histological methods such as Perls' or Turnbull staining are the best known techniques. Perls' stain reacts mainly with Fe³⁺ ions, giving an intense Prussian (or Berlin) blue. However, it also reacts with Fe²⁺, resulting in a white precipitate known as Everitt's salt that slowly oxidizes to give Prussian blue (Meguro et al., 2007). Perls' staining only makes it possible to highlight non-heme iron. The iron present in the heme has strong covalent bonds and only treatment with strong alkalis (ammonium sulfide) can dissociate it (Meguro et al., 2007). By contrast, the iron bound to iron metabolism molecules (ferritin, hemosiderin, transferrin, iron-sulfur proteins, etc.) can easily be ionized from a pH of around 4.5. This explains why treatment with hydrochloric acid is necessary for Perls' staining. Like iron heme, iron present in neuromelanin and probably lipofuscin is historically described as invisible iron, owing to its strong affinities. As a result, it cannot be revealed by Perls' staining. The second best known histochemical method is the Turnbull technique, which consists in applying potassium ferricyanide instead of potassium ferrocyanide, again in association with hydrochloric acid. This treatment makes it possible to reveal Fe²⁺ ions. Other methods involve using the iron-sulfur reaction to demonstrate both Fe²⁺ and Fe³⁺. Treatment with ammonium sulfide reveals Fe²⁺ (Quincke method), and pretreatment with ammonium sulfide and secondary treatment with Turnbull stain reveal Fe³⁺ in addition to Fe²⁺ (Tirmann-Schmelzer method). The Tirmann-Schmelzer method is an iron-sulfur method. Theoretically, the iron-sulfur technique has three drawbacks. First, not all Fe³⁺ is converted to Fe²⁺, thus limiting the sensitivity of the technique. Second, ammonium sulfide treatment also extracts heme iron in theory but only if the treatment is at particular low pH and/or for a very long period (several days) compare to classical protocols (90 min). Third, sulfur-based methods are not specific for iron and causes the precipitation of other metal ions, such as copper or zinc into copper sulfide or zinc sulfide. However, these precipitates do not further react with potassium ferricyanide (Turnbull reaction). So, while the simple iron-sulfur techniques react with zinc and copper, the actual Tirmann-Schmelzer method does not suffer from this lack of specificity, because the specificity for iron comes from the reaction of Fe²⁺ ions with potassium ferricyanide (Meguro et al., 2007). The Tirmann-Schmelzer method has recently been used to study iron in rigorous studies with good results (Wiggermann et al., 2017; Hametner et al., 2018).

To enhance the visualization (augmented sensitivity) of Perls' or Turnbull staining, diaminobenzidine (DAB) can be added, turning Perls' or Turnbull's blue to a brown palette (cf. Figure 1). Another colorimetric assay method involves reacting ferrozine with Fe²⁺ to give a magenta red precipitate that can then be assayed spectrophotometrically. Pretreatment with hydroxylamine hydrochloride converts Fe³⁺ to Fe²⁺.

Although the iron itself cannot be revealed by immunohistochemistry, the same is not true of the proteins containing this iron. The advantage of this technique is that it identifies antigens (generally proteins) of interest at the cellular or extracellular level using specific antibodies. These antibodies may be of either polyclonal or monoclonal origin. Polyclonal antibodies recognize various antigens of the protein of interest. There is therefore a risk of cross-reactions. Monoclonal antibodies recognize a single antigen. They have greater specificity, but diminished sensitivity.

The main proteins that can be targeted are transferrin, FP, iron-responsive proteins, and ferritin. There are both polyclonal and monoclonal antibodies for ferritin. However, each antibody is specific for either the L chain, the H chain, or mitochondrial ferritin.

In conclusion, there are no perfect staining methods for accurately calculating the iron concentration of a given tissue. They are therefore considered to be semiquantitative methods at best. Their strength lies in their qualitative character to highlight areas rich in iron with an excellent spatial resolution, whether macroscopically or microscopically.

Physical techniques like inductively coupled plasma mass spectrometry (ICP-MS) or atomic absorption spectroscopy (AAS) are used to quantify iron concentrations with precision, but they cannot map the iron in the brain. Recent techniques such as laser ablation inductively coupled plasma mass spectrometry (LA-ICP-MS) or X-ray fluorescence (XRF) can robustly quantify and precisely map iron in brain structures (Matusch et al., 2011; Zheng et al., 2013).

Ferritin, hemosiderin and deoxyhemoglobin are considered to be the main forms of iron present in sufficient quantities to affect MR contrast in the human brain (Schenck, 2003; Haacke et al., 2005). To a lesser extent, neuromelanin can also affect the MRI signal in particular locations. Lipofuscin's signal impact needs more investigation. Proteins associated with small numbers of iron atoms like transferrin are often regarded as negligible. Consider the case of ferritin. It is not so much its total composition that is important as its core, most often a ferrihydrite crystal which, like all metallic nanoparticles, has magnetic properties capable of affecting susceptibility. Ferrihydrite is an antiferromagnetic structure with, by definition, antiparallel magnetic ordering with moment compensation. The magnetization vector in a crystal is not isotropic, as the direction of spin relative to the crystal lattice depends on magnetocrystalline anisotropy. This anisotropy stems from the presence of easy magnetization directions within the crystal. However, the nanometric size of the core (about 6 nm) gives it different properties. Below a certain size, particle magnetization relaxes above the energy barriers of anisotropy. If the volume of the particle is small enough, the energy barriers separating the minimum energy directions (easy directions) may correspond to thermal energy at room or body temperature. The magnetic moment can fluctuate from one easy direction to another. This thermally activated phenomenon constitutes superparamagnetic relaxation (Daou, 2007; Bennett et al., 2008).

It should also be noted that the particle's magnetic moment depends on the number of iron atoms and surface/volume ratio in ferritin. The maximum capacity is 4,500 atoms, but the actual capacity often oscillates around 2,000 atoms. Finally, the crystal is not always in the form of ferrihydrite, as it can also be in the form of magnetite/maghemite, which has a magnetic moment 10 times greater than the crystal of ferrihydrite.

Magnetic susceptibility, denoted as chi (χ), is defined as the ratio of the magnetic moment (M) to the magnitude of the external magnetic field (B). This variable is dimensionless.

Its molar magnetic susceptibility (χm) is equal to its magnetic susceptibility (χ) multiplied by the molar mass of the test compound.

Diamagnetic substances have a molar susceptibility of <1 ppm. This is the case of most human tissue, which is very weakly diamagnetic. By contrast, paramagnetic and superparamagnetic substances have a molar susceptibility >1 ppm, and some substances, such as ferritin, have a very high molar susceptibility (5,000 ppm).

In other words, highly (super- or ferro-) paramagnetic substances significantly increase the magnetic field at the local level. This increase in the magnetic field results in changes in the frequency of spin precession, and thus in a poor location of the MR signal (image distortion). These local inhomogeneities of the magnetic field also have the effect of shortening the T2^* or T2 times, by accelerating the spin phase shifts in between.

This is called the susceptibility artifact. Magnetic susceptibility, phase shift and shortening of T2 ^* are the main approaches to the quantification of iron in MRI.

Advances in MRI mean that we are now able to estimate the presence of iron in the brain. Several techniques have emerged to quantify iron content. Some are based on transverse relaxation time (T2, T2^* and T2') or its inverse (R2, R2^* and R2'), in the relation R2^* = R2 + R2'. Iron, especially when it is stored in ferritin or hemosiderin, induces local magnetic field inhomogeneities and increases R2 and R2^* relaxation rates. Transverse relaxation rates (R2 or R2^*) can be measured by multi-echo acquisition (spin echo for R2 and gradient echo for R2^*), then voxelwise fitting to an exponential model (Péran et al., 2007).

The second type of sequences used to estimate iron are based on paramagnetic effects from iron-containing proteins that increase bulk tissue magnetic susceptibility. Magnetic susceptibility information is contained in phase images. Susceptibility weighted imaging (SWI), which combines phase images with magnitude images to enhance contrast among structures with different local susceptibilities, was the first to be used. Quantitative susceptibility mapping (QSM) was developed more recently to directly measure susceptibility and be independent of imaging parameters such as echo time or field strength (de Rochefort, 2010; Deistung et al., 2016). QSM is based on phase images and uses complex post-treatment algorithms to remove background field components after phase unwrapping. Several algorithms and methods can be used, such as Laplacian-based methods, calculation of susceptibility using multiple orientation sampling (COSMOS), morphology-enabled dipole inversion (MEDI), and sophisticated harmonic artifact reduction on phase data (SHARP) method (Wang and Liu, 2015; Deistung et al., 2016). QSM allows investigators to distinguish between calcium and iron, unlike R2^*. There are other sequences, such as field-dependent R2 increase (FDRI), that have high sensitivity and specificity for iron, but they are seldom used in clinical practice, owing to technical constraints.

Iron-based sequences are used in research or clinical routine in healthy participants (Péran et al., 2009) to estimate the amount of iron in brain structures, especially the basal ganglia, or to compare patients with diseases such as PD with healthy controls (Péran et al., 2010) or establish differential diagnoses for disorders such as MSA (Barbagallo et al., 2016; Péran et al., 2018).

Researchers trying to correlate iron-based sequences with an estimation of iron concentration in brain structures draw on data from a seminal work published in 1958 by HALLGREN and SOURANDER (1958) and verified in recent research (Krebs et al., 2014), using regression line equations for many parts of the brain, where y = concentration of non-heme iron per 100 g fresh weight and x = age of participant. For all types of iron-based sequences [FDRI Bartzokis et al., 2007,R2^*Péran et al., 2009, or QSM Bilgic et al., 2012], correlations are calculated with Hallgren and Sourander' s equations.

Despite these rigorous data, the desire to correlate estimations of iron with MRI using a reference quantitative technique has led scientists to undertake ex vivo MRI, which allows for further investigations after scanning. Ex vivo MRI has many advantages compared with in vivo MRI, including longer acquisition time for better spatial resolution, air-tissue interface elimination, no importance of the specific absorption rate (SAR), no patient artifacts, and the possibility of using a dedicated receiver coil for small anatomical specimens. Ex vivo MRI has gained in popularity in the scientific community, not only for anatomical atlases and forensic medicine, but also for molecular and trace element mapping.

The aim of the present literature review was to systematically describe and evaluate published studies that used ex vivo MRI combined with histological or physical techniques to detect iron in the brain. We focused on quantitative studies that attempted to find statistical correlations between the two kinds of techniques, but also included articles exploring iron in relation to a specific pathology or just for anatomical description purposes. After summarizing their results, we discuss all the pitfalls that can be encountered along the way.

The initial search was undertaken on PubMed using the following terms: “brain” AND “magnetic resonance imaging” OR “MRI” OR “MR” OR “QSM” OR “R2^*” AND “Ex-vivo” OR “post-mortem” OR “postmortem” AND “iron.” The review was carried out in accordance with PRISMA guidelines for systematic reviews (Moher et al., 2009). Our search criteria yielded 486 results. We limited the articles we reviewed to human studies and papers written in English. Relevant papers were identified by manual review of titles and abstracts. Articles with no evidence of iron staining or detection by a physical technique on the MR scanned specimen were excluded. A total of 51 articles were fully analyzed, and two were excluded owing to the use of proton relaxation measurements rather than MRI (Vymazal et al., 1996; Ye et al., 1996). We distinguished quantitative analyses from qualitative (visual assessment) ones in the 49 final articles and listed all the different pathologies concerned (Figure 2). Relevant data were listed. In the case of the quantitative studies, we identified the linear regression parameters, along with their linear equation parameters. Finally, key methodological points were noted, in order to highlight limitations that may be encountered when matching ex vivo MRI with histology.

Table 1 sums up all the recent literature on post mortem MRI and histological or physical techniques for measuring iron. Papers were classified as quantitative (25 articles) when correlations were sought between MR data and histological or physical techniques, and qualitative (24 articles) when images were simply visually assessed. Where articles described two reference techniques (often a histological one and a physical one), the quantitative assessment was only applied to the physical technique. In some papers, quantitative measures were carried out in order to compare data between the pathological group and controls, so they were not classified as quantitative.

MR scanners varied widely in main field strength (1.4–21.1 tesla), with the largest, supported by micro-MRI, being particularly useful for small specimens. Many MRI techniques were used to highlight iron, but two recent techniques (R2^* and QSM) seemed particularly valuable. A total of 17 articles published after 2009 featured R2^* techniques (Yao et al., 2009; Fukunaga et al., 2010; Langkammer et al., 2010; Bagnato et al., 2011, 2018; Antharam et al., 2012; Kwan et al., 2012; Langkammer, 2012; Walsh et al., 2013; Stüber et al., 2014; Dusek et al., 2015; Meadowcroft et al., 2015; Wisnieff et al., 2015; Birkl et al., 2017; Bulk et al., 2018b; Hametner et al., 2018), while 10 published after 2012 used QSM techniques (Langkammer et al., 2012; Zheng et al., 2013; Stüber et al., 2014; Tan et al., 2014; Sun et al., 2015; Wisnieff et al., 2015; Wiggermann et al., 2017; Hametner et al., 2018; Lee et al., 2018; Kaunzner et al., 2019). However, many more techniques (T1 mapping, T2, T2 mapping, R2, R2', SWI, phase, etc.) were linked more or less successfully with iron. Some authors worked on the relation between transverse relaxation time (R2^* and R2') and specimen temperature, developing an iron quantification method based on a temperature coefficient (Birkl et al., 2015, 2017). This seems a very interesting approach to investigating the relationship between MR signal, iron, myelin and temperature, and is a good example of what can be done with ex vivo material compared with in vivo. This technique now needs to be reproduced by other teams to confirm its effectiveness.

The types of anatomical specimens used differed greatly. Some studies used excisional material from surgery, such as gangliogliomas or cavernomas, while others compared pathological samples from a brain bank with healthy controls, or fresh specimens from required autopsies or anatomy laboratories. Anatomical specimens also differed in size, with some teams using the whole brain, many others only a thick coronal slice (often from brain banks) and the remainder using a small anatomical part, such as the substantia nigra, putamen, subthalamic nucleus, or medium temporal gyrus.

Regarding pathologies, AD (11 articles), MS (9 articles), PD (3 articles), progressive supranuclear paralysis (PSP) (3 articles), MSA (2 articles) and amyotrophic lateral sclerosis (2 articles) were the most frequently studied conditions.

Regarding the physical techniques used for iron quantification, ICP-MS was the one most often used (6 articles), with LA-ICP-MS performed in one study (Wisnieff et al., 2015). The second most frequent technique (4 articles) allowing for mapping was XRF. An interesting technique based on the substance's magnetic characteristics, known as superconducting quantum interference device magnetometry (SQUID), was used to further characterize the paramagnetic properties of the particles, allowing, for example, ferrihydrite crystal to be differentiated from magnetite/maghemite (Birkl et al., 2015; Bulk et al., 2018b).

Given that physical techniques require expensive materials, physicians, expert technicians and sound knowledge of physics, histology remains of interest because of its low cost, simplicity and opportunities for mapping. Perls' staining was the most frequently used technique (19 articles for Perls and 8 articles for Perls + DAB), followed by Turnbull + DAB + ammonium sulfide (5 articles).

Table 2 sums up all the different combinations of MRI and histological or physical techniques.

Although correlations between in vivo MRI (especially R2^* and QSM) and Hallgren's equations yielded remarkably reproducible results across the different teams (Ghassaban et al., 2018), many centers tried to confirm these by correlating MRI and histological/physical data. We selected 25 articles because one of the goals of the studies they described was to quantify iron using a histological/physical technique and correlate it with MRI sequences. All these studies are summarized in Table 3. Relevant MRI sequence acquisition parameters are provided. When available, Pearson (r) or Spearman (ρ) correlation coefficients or coefficients of determination (r²) are given in Column 6 with their associated p-value. Linear equation parameters are set out in Column 7. As only relevant or overall results are given, they do not reflect all the correlations found in the studies.

Quantitative correlations have been attempted with many physical techniques, but advances in digitization and image post-processing software have recently led to a renewal of interest in Perls' or Turnbull staining. Table 4 lists all the articles describing MRI + Perls' or Turnbull staining. Staining techniques and slice thickness are provided, along with the digitization step and a brief description of the correlation methodology. Visual always refers to qualitative assessment.

We reviewed all the relevant literature concerning ex vivo MRI and iron histology and came to three main findings. First, matching ex vivo MRI with histology allows for a better understanding of the physiopathology of some neurological disorders in which iron is suspected to play a role, as well as of the radiological anatomy of specific neurological structures. Second, correlating iron-based MRI sequences with iron quantification performed with a reference physical technique shows that MRI sequences can be reliably used for the in vivo estimation of iron in the brain. It is interesting to note that results from ex vivo MRI were in accordance with in vivo data correlated using Hallgren and Sourander's equations. Third and last, data from ex vivo MRI correlated with quantitative staining seem to be a good approximation of the results yielded by physical techniques, and histology has the added advantages of being relatively inexpensive and offering opportunities for mapping. However, matching ex vivo MRI and histology raises some methodological issues that need to be taken into account before extrapolating to in vivo MRI.

No datasets were generated or analyzed for this study.

AD, GA, and PP contributed conception and design of the review. AD and JC wrote the first draft of the manuscript. GA wrote sections of the manuscript. AD, GA, and JC collected data. PC reviewed specifically histological and anatomical parts. PP reviewed specifically MRI parts. All authors contributed to manuscript revision, read and approved the submitted version.