The value of magnetic resonance imaging as a biomarker for amyotrophic lateral sclerosis: a systematic review

Background Amyotrophic lateral sclerosis (ALS) is a fatal, rapidly progressive neurodegenerative disease that mainly affects the motor system. A number of potentially neuroprotective and neurorestorative disease-modifying drugs are currently in clinical development. At present, the evaluation of a drug’s clinical efficacy in ALS is based on the ALS Functional Rating Scale Revised, motor tests and survival. However, these endpoints are general, variable and late-stage measures of the ALS disease process and thus require the long-term assessment of large cohorts. Hence, there is a need for more sensitive radiological biomarkers. Various sequences for magnetic resonance imaging (MRI) of the brain and spinal cord have may have value as surrogate biomarkers for use in future clinical trials. Here, we review the MRI findings in ALS, their clinical correlations, and their limitations and potential role as biomarkers. Methods The PubMed database was screened to identify studies using MRI in ALS. We included general MRI studies with a control group and an ALS group and longitudinal studies even if a control group was lacking. Results A total of 116 studies were analysed with MRI data and clinical correlations. The most disease-sensitive MRI patterns are in motor regions but the brain is more broadly affected. Conclusion Despite the existing MRI biomarkers, there is a need for large cohorts with long term MRI and clinical follow-up. MRI assessment could be improved by standardized MRI protocols with multicentre studies.


Background
Amyotrophic lateral sclerosis (ALS) is a neurodegenerative disease that mainly affects the motor system. At present, the only drug to have produced an increase in patient survival in a controlled clinical trial is riluzole [1]. The condition is always fatal; the median survival time after onset is 36 [2], although there are great individual variations [3]. Although ALS is a clinically recognizable condition (in terms of the pattern of progressive upper and lower motor neuron degeneration), the clinical presentation and progression are heterogeneous [4]. Moreover, there are now strong reasons for considering a broader aetiological and pathogenic spectrum [5]. This clinical heterogeneity is observed in other neurodegenerative diseases, such as Parkinson's disease (PD). Nevertheless, it has been found that all PD patients display degeneration and iron overload of the substantia nigra, which might become a surrogate biomarker [6]. In ALS, iron overload in the motor cortex was confirmed [7] after being suspected in mouse models [8]. Ongoing drug trials in the field of neurodegenerative disease are mainly seeking to establish neuroprotection and neurorestoration. Since neuroprotection can never be unambiguously demonstrated in a given patient, the concept of disease-modifying drugs (with a slowing of the disease progression) has arisen. At present, the evaluation of a drug's clinical efficacy in ALS is based on the ALS Functional Rating Scale Revised, motor tests, survival or a combination of these measures (such as the Combined Assessment of Function and Survival (CAFS)) [9,10]. However, these endpoints are general, variable and late-stage measures of the ALS disease process and thus require the long-term assessment of large cohorts. These assessments are risky and expensive, which considerably limits the number of trials being conducted. A sensitive surrogate biomarker might help to (i) reduce the sample size in pilot studies, (ii) better define the sample size required in Phase III clinical trials (by comparison with clinical scales) and (iii) define endophenotypes for separate assessment in clinical trials.
Biomarkers are typically divided into "wet biomarkers" and "dry biomarkers". Wet biomarkers are biological substances that are measured in a body fluid (such as whole blood, serum, plasma, saliva or cerebrospinal fluid (CSF)). "Dry biomarkers" are based on functional scales, task performance, electrophysiology or imaging. Magnetic resonance imaging is a widely available and non-invasive technique, which makes it an appropriate candidate for biomarkers. Although early reports stated that ALS starts in the spinal cord, the initial site of neurodegeneration has not been unambiguously identified; disease may start in the spinal cord, in the motor cortex or at both sites simultaneously [11,12]. The development of radiological biomarkers in ALS is facilitated by the hypothesis that the pathological changes in ALS radiate out from the initial spinal and brain areas in which the most severe upper and lower motor neuron dysfunction respectively occur (from symptom onset onwards) with variability in the predominance of upper and lower motor neuron dysfunction [13]. The neurodegeneration then progresses along the motor neurons of the ventral horn of the spinal cord and into the corresponding areas of the motor cortex [13]. Despite this modelization, the pattern of onset and progression remains complex [13]. Hence, MRI of the spinal cord should be an early, sensitive marker of subtle degenerationrelated changes. However, MRI of the spinal cord is complicated by a number of technical difficulties (relative to MRI of the brain). Firstly, the magnetic field is inhomogeneous because the spinal cord is close to bone, soft tissues and air. Secondly, the spine has a small cross-sectional dimension, which thus requires high-resolution data acquisition. Thirdly, the spine is subject to motion of the CSF [14] and physiological motion due to breathing [15].
Many different MRI techniques and sequencesincluding voxel-based morphometry (VBM), diffusion tensor imaging (DTI), magnetic resonance spectroscopy (MRS), iron-sensitive sequences (T2*, R2* and SWI) and functional MRI (fMRI) -are available for studying ALSrelated changes in the brain or spinal cord. Hence, we reviewed the correlations between MRI findings and clinical scores in ALS, and sought to establish whether particular sequences may have value as surrogate biomarkers in this disease.

Methods
We searched the PubMed database with the following keywords: "ALS AND MRI OR amyotrophic lateral sclerosis AND magnetic resonance imaging", "ALS AND VBM OR amyotrophic lateral sclerosis AND magnetic resonance imaging ", "ALS AND MR spectroscopy OR amyotrophic lateral sclerosis AND magnetic resonance spectroscopy", "amyotrophic lateral sclerosis AND diffusion tensor imaging OR ALS AND DTI", "amyotrophic lateral sclerosis AND functional magnetic resonance imaging OR ALS AND fMRI". There was no limitation on the publication date but only publications in French or English were considered. We included (i) general MRI studies with a control group and an ALS group and (ii) longitudinal follow-up studies (even if a control group was lacking). We excluded therapeutic trials articles, reviews, studies focused only on particular subgroups of ALS patients such as cognitive impairment, genetic mutations.

Results
A total of 1336 articles were identified using database searching, 651 were recorded after duplicates removal. 405 were excluded (not ALS, not MRI or, therapeutic clinical trials, not English or French, reviews) and 130 articles were finally excluded (studies focused only on particulars subgroups of patients, other MRI sequences or lacking of a control group). We analysed 116 articles.

Brain imaging
Structural magnetic resonance imaging (Table 1) Three methods based on a three-dimensional, T1weighted sequence have been applied to ALS: VBM, volume analysis and the measurement of cortical thickness. As a volume-based analysis, VBM involves the automatic segmentation of grey matter and inter-subject comparison of the local grey matter density [16,17]. After all the data have been transformed in the same stereotaxic space, the images are partitioned and corrected for the separation between white matter, grey matter and CSF as a function of voxel intensity [17]. Measurement of cortical thickness is a surface-based technique analysis in which the white matter and pial surfaces are automatically extracted by applying image analysis packages such as Freesurfer (http://surfer.nmr.mgh.harvard.edu/).
Clinical correlations Specific MRI abnormalities have been correlated with clinical scores, the disease stage (as estimated with the original or revised ALS Functional Rating Scale (ALSFRS and ALSFRS-R, respectively), the speed of disease progression (i.e. the change in ALSFRS or ALSFRS-R score, divided by the disease duration or the time interval between two clinical assessments), the disease duration and the survival time. Lower ALSFRS or ALSFRS-R scores were associated with larger ventricles and greater volume loss in the basal ganglia [38], and volume changes in several parts of the frontal lobe (such as Brodman area 10) [18]. In another study, a higher ALSFRS-R progression rate was associated with MRI changes in the left sensorimotor area; cortical thickness was lower in patients with predominantly upper motor neuron impairment (especially in the precentral gyrus and the left paracentral lobule), whereas patients with predominantly lower motor neuron impairment showed less difference relative to controls [20]. Grey matter density was lower in the cingulate and right inferior frontal gyrus in patients with impaired emotional empathy [36]. Poor survival in ALS is correlated with volume loss in the basal ganglia and limbic structures [38]. The initial side of limb weakness in ALS patients was not correlated with contralateral brain cortical loss of volume [27]. Although atrophy was predominant in the left sensorimotor cortex [27,28], it was not related to the side of onset [27]. A subgroup analysis (by side of onset) found more widespread atrophy in the contralateral brain cortex than in the group of ALS patients as a whole [28]. Precentral cortical thickness was lower in patients with upper motor neuron clinical features than in those with lower motor neuron clinical features and in the ALS group as a whole [29].
In a study with a post mortem T1 sequence and histological assessment, an abnormally small difference in intensity between cortical grey matter and subcortical white matter was observed not only in the motor cortex (including the precentral gyrus, supplemental motor area and premotor cortex) but also in somatosensory areas and the primary visual cortex. These signal abnormalities were linked to neuron loss and an elevated number of astrocytes [41].
Diffusion tensor imaging (Table 2) DTI (also known as magnetic resonance tractography) is based on the random diffusion (Brownian motion) of molecules. In a spherical volume, the diffusion of water has no main direction and its diffusion in the three directions (λ1, λ2 and λ3) is equally likely. However, this is no longer the case in an ellipsoid or cylindrical volume and the diffusion is anisotropic. λ1 is the volume's main axis (axial diffusivity, AD) and λ2 and λ3 are the minor axes (radial diffusivity, RD). By analysing the diffusion of water in three directions, four parameters can be defined: FA, RD, AD and mean diffusivity (MD, which is the average of diffusion in the λ1, λ2 and λ3 axis, also referred to as the apparent diffusion coefficient (ADC)). The first three parameters describe the spatial variation Agosta et al., [21] 25/18 GM volume reduction in right precentral, left inferior frontal cortex and temporal superior cortex.
Negative with ALSFRS-R and DD.
DPR is correlated with the mean cortical thickeness of left sensorimotor cortex.
Not available.
Canu et al., [23] 23/24 GM volume reduction in the precentral, right opercular and angular cortex, WM volume reduction in the frontal lobe (especially the subcortical motor areas) and the temporal lobe.
Not available.
Not available.
Chang et al., [22] 20 ( Compared to limb onset ALS, bulbar onset ALS showed GM volume reduction in the brainstem, the cerebellum and in the fusiform gyri (BA 34) and WM volume reduction along the left corticospinal tract.
ALSFRS-R correlated positively with GM volume reduction in the right medial frontal gyrus (BA 10).
Kassubek et al., [32] 22/22 GM volume reduction in the right primary motor cortex and the left medial gyrus and bilateral inferior temporal gyrus.
Negative with ALSFRS-R and DD.
Meadcroft et al., 2014 Post mortem study; loss of signal difference between WM and GM (mainly in the primary motor cortex).
Not available.
Not available.
Mezzapesa et al., [20] 29/20 GM volume reduction in the precentral, frontal, right temporal and right occipital cortex. Thivard et al., [24] 15/25 GM volume reduction in the precentral, frontal and temporal cortex (especially the hippocampus), the parietal and occipital cortex, the thalamus and the cerebellum. No difference in WM.
Not available.
Verstraete et al., [30] 12/12 GM reduction in the right and left precentral cortex. Not available.
of water movement and are related to the orientation of the studied structures. In contrast, MD corresponds to the mean displacement of the water molecules within the volume. The main application of DTI is the analysis of the anatomical bundles that compose the white matter of the brain and the spinal cord [42][43][44]. Whereas DTI corresponds to the analysis of the above-mentioned parameters in a defined volume, tractography corresponds to the analysis of a complete white matter tract.
A tractography study revealed that the CST volumes of ALS patients are lower than in controls [71]. A DTI study using a tract of interest-based fibre tracking found that the radiological results mirrored the neuropathological staging system [72], suggesting that the disease process extended in the following sequence: the corticospinal tract (stage 1), the corticorubral and corticopontine tracts (stage 2), the corticostriatal tract (stage 3) and the proximal portion of the perforant pathway (stage 4) [73]. The researchers suggested that this MRI-based staging system could be used as a surrogate marker of disease progression in clinical studies of ALS.
Radial and axial diffusivity in ALS Several studies have found elevated RD and AD values in the right CST, the right cingulum and the left anterior thalamic radiations (on the basis of changes in FA or MD) [53]. In the PLIC, AD is low [63] and RD is elevated [67] -especially in patients presenting with T2 hyperintensities in CSTs [63]. RD is elevated in white matter connected to the primary motor cortex, the premotor cortex [56] and the corpus Table 1 The main results of volumetric studies (Continued) Walhout et al., [29] 153 (112 ALS, 19 UMN, 19 LMN)/60 GM reduction in the right and (predominantly) left precentral cortex and the right paracentral cortex.
Bulbar and arms ALSFRS-R subscores are correlated with cortical thickness of corresponding precentral cortex areas of the motor homunculus. DPR is correlated with the thickness of the right inferior temporal cortex, the postcentral cortex and the right paracentral cortex.     [56,67,69]. Lastly, RD and AD are elevated in the cerebellum of C9ORF72-positive patients [75].
Longitudinal studies In longitudinal analyses, low FA is observed in the right superior CST [62], precentral subcortical regions, mesencephalic CSTs and parts of cerebellum [61]. FA was found to fall in a progressive, linear manner along the CSTs [34,46,76], and the diffusivity of the external and internal capsule was elevated [61]. AD is elevated in CSTs [77]. However, a recent longitudinal study found that white matter defects (as assessed by DTI) progressed much slowly that in the grey matter defects -notably in the basal ganglia [78].
Clinical correlations Many studies have found that disease progression (as defined by the change over time in the ALSFRS or ALSFRS-R score) is correlated with the changes in FA in CSTs [24,34,45,61,62,77,79]. Furthermore, the disease duration is correlated with FA in entire CSTs or parts of CSTs [24,50,54,56,61,77]. Conversely, the disease duration was negatively correlated with FA in the cerebellum [61], the subcortical white matter of insula, the ventrolateral premotor cortex, the cingulum, the precuneus and the splenium of the corpus callosum [24]. Lower-limb muscle strength was correlated with the contralateral FA in CSTs [60]. Diffusivity values in CSTs and in the parietal lobe, temporal lobe and cerebellum were also correlated with disease progression [52,79]. Diffusivity in CSTs was correlated with disease duration [79], and contralateral diffusivity in CSTs was correlated with spasticity [59]. Whereas both limb-onset patients and bulbar-onset patients differed from controls in terms of FA along CSTs, the two patient groups had similar values [60]. The CSTs were more impaired in bulbar-onset patients than in limb-onset patients when considering FA, MD and RD but not AD [80]. Tractography revealed a large number of differences between controls and patients defined as definite/probable according to the El Escorial criteria, whereas there were fewer differences between controls and a possible/suspected group [79].
Diagnostic accuracy An individual patient data metaanalysis of CST studies concluded that DTI lacks sufficient diagnostic discrimination [81]. The pooled sensitivity and specificity were only 0.68 and 0.73, respectively. Researchers have also suggested that a multimodal approach (combining DTI with methods such as MRS) may be more promising [55].
Magnetic resonance spectroscopy (Table 3) MRS is a non-invasive means of assessing brain metabolism, based on the chemical properties of molecular structures. Results are expressed as peaks. The main brain metabolites monitored with MRS include Nacetyl-aspartate (NAA), choline (Cho, a cell membrane marker), creatine (Cr) and phosphocreatine (Pcr). The Cr + Pcr peak is a marker of energy metabolism; the level of Cr is assumed to be stable and is used to calculate the metabolite ratio. At a field strength of 1.5 T, the peaks for glutamine (Gln), glutamate (Glu, an excitatory neurotransmitter) and gamma aminobutyric acid (GABA, an inhibitory neurotransmitter) overlap to form a complex referred to as "Glx"; higher field strengths are required for more accurate quantification of this metabolites. The simple sugar myo-inositol (Ins) is absent from neurons but present in glial cells. Hence, elevated brain levels of Ins are associated with glial proliferation, whereas low levels are associated with glial destruction [82][83][84].
The main results for the CST concern the ratio between the NAA peak on one hand and the Cho and Cr + Pcr peaks on the other. In ALS patients, low levels of NAA have been observed in the precentral corticosubcortical region [85][86][87][88][89][90][91], in the PLIC [85] and in entire CSTs [55,59,86,92]. Glu and Glu + Gln were found to be elevated in the precentral corticosubcortical region and in the PLIC [85], whereas Cho was found to be elevated in the precentral region [88,93,94] and along the CSTs [47,59,92]. Levels of the inhibitory neurotransmitter GABA were low in the precentral cortex [55,95], and levels of Ins were low in the precentral corticosubcortical region [89].
Whereas the above-mentioned studies used defined voxels for MRS acquisition, whole-brain spectroscopy has revealed low NAA/Cho or NAA/Cr ratios in many other regions [93].
Follow-up studies have shown a decrease in the NAA peak in the precentral cortex (with no differences for other metabolites) [96], a decrease in the NAA/Cho peak and an increase in the Cho/Cr peak [91].
Clinical correlations Patients with definite or probable ALS (according to the El Escorial criteria) had low levels of NAA and high levels of Ins and Cho in the precentral cortex. In patients with possible or suspected ALS, only Cho and (on the left side) Ins were elevated. In another study of suspected ALS patients, only right-side Ins and right-side Cho were elevated [79]. Low NAA and high Cho and Ins levels are associated with clinical disease severity [94]. Lower NAA levels in the precentral gyrus, disease duration and disease progression were intercorrelated [59], and a low NAA/Cho ratio was associated with poor survival [88].
Diagnostic accuracy In a recent study, the combination of MRS and DTI yielded a sensitivity of 0.93 and a specificity of 0.85, whereas the DTI data alone gave values of 0.86 and 0.70, respectively [55]. A criterion combining the MRS data, hypointensity in the precentral gyri and hyperintensity in the CSTs yielded a sensitivity of 0.78 and a specificity of 0.82; when each of the three parameters was considered alone, the sensitivity and specificity values did not exceed 0.71 and 0.75, respectively [90].

Iron imaging
Iron overload appears to be involved in the physiopathology of ALS [8,97]. However, very few studies have analysed the iron content in the brain of ALS patients. A variety of techniques are available, including relaxometry (T2, T2* or R2*) and the susceptibility-weighted imaging [98,99] (SWI, which is probably more sensitive than relaxometry [100]). Multi-echo T2* and R2* sequences are sensitive to magnetic field inhomogeneities. The value of R2* is measured by voxel-by-voxel modelling of the exponential decrease in signal [101,102]. In postmortem studies, R2* values have been found to be correlated with the iron content [103].

Susceptibility-weighted imaging and quantitative susceptibility mapping (QSM)
SWI is derived from T2* sequences and can be described as a flow-compensated, high-spatial-resolution T2* weighted sequence. The main advantage over conventional T2* sequence is better contrast [101]. QSM is based on the analysis of phase information. The main sources of phase shifts in biological tissues are the iron, calcium, lipid and myelin contents [104]. The value of QSM in ALS was confirmed in a post-mortem study, in which the iron content of grey matter was assessed more accurately than that of white matter [105]. The initial studies in the field noted a shorter T2 time in the motor cortex of most ALS patients (vs. controls) [106], although this was not confirmed a few years later [107]. Later studies found a correlation being T2* shortening and iron deposition in the microglia of the motor cortex (according to a post-mortem examination) [7,106,108]. Iron overload in the motor cortex was also described in SWI studies [109] and qualitative studies [100]; in the later study, there were no differences between ALS patients and controls in terms of T2 or T2*. A few quantitative studies have looked at white matter or deep grey matter [109,110]. There was a trend towards iron accumulation in CST white matter in a relaxometry study [110]. Iron overload was also present in the red nucleus, substantia nigra, globus pallidus, putamen [109] and caudate nucleus [110]. The use of SWI sequences revealed widespread iron overload and myelin defects in the white matter of all brain lobes and corpus callosum [111]. A retrospective study of iron deposition in motor cortex found that quantitative susceptibility mapping was more accurate than T2*, T2 or FLAIR sequences [112].
A follow-up study has highlighted the progression of a T2 hypointense area in the motor cortex after 6 months, which was strongly and negative correlated with the ALSFRS score [108]. Lastly, changes in SWI sequences in the corpus callosum are correlated with the ALSFRS-R score [111].

Imaging of the spinal cord
The few studies to have analysed the spinal cord revealed atrophy in the cervical and upper thoracic regions [113][114][115]. A DTI study reported low FA, elevated RD and a high magnetization transfer ratio in lateral segments of the cervical spinal cord [114]. Atrophy of the cervical and upper thoracic spinal cord progressed over time, whereas the DTI changes were relatively stable [115]. This observation agrees with the results of a longitudinal brain MRI study showing that grey matter defects (but not white matter defects) progressed over time [78]. Sensory pathways were also involved, with a decreased FA and an elevated MD and elevated RD in the posterior part of the cervical spinal cord [115,116]. MRS of the cervical spinal cord revealed low NAA/Cr + Pcr and NAA/Ins peaks and an elevated Ins/Cr peak [117].
Clinical correlations Cervical and upper thoracic spinal cord atrophy was correlated with upper arm muscle strength, as measured by manual testing [114]. FA in the cervical spinal cord was correlated with the ALSFRS score. The rate of atrophy in a follow-up study was correlated with the rates of changes of an arm ALSFRS-R score and an arm strength score (in a manual test) [115]. The NAA/Cr + Pcr and NAA/Ins peaks were positively correlated with the ALSFRS-R score and forced vital capacity but negatively correlated with disease progression [117].

Functional MRI
Brain fMRI is a recent technique based on blood flow and blood-oxygen-level dependent (BOLD) contrast, which in turn are based on the neurovascular consequences of neuronal activation [118]. The paramagnetic properties of deoxyhaemoglobin lead to signal reduction in gradient echo sequences [101]. Brain fMRI sequences are acquired in the resting state or during performance of a motor, auditory, cognitive, or visual task [119][120][121]. The resting state sequences are quite similar to T2* sequences [101].
In the resting state, functional connectivity is based on fluctuations of the BOLD signal. In general, seven "default" networks can be identified: default mode, executive control, visual salience, sensorimotor, dorsal attention and auditory networks [118,122]. Functional connectivity between specific brain regions and/or networks and the rest of the brain can then be analysed [118]. Resting-state analysis can also provide information about local activation (by analysing the amplitude of low-frequency fluctuations (ALFF), for example).
The resting state Resting-state studies have evidenced the widespread reorganization of functional connectivity in ALS patients. This reorganization can be observed as either elevated functional connectivity or low functional connectivity. Functional connectivity was found to be low in several parts of the frontal lobe (the right orbitofrontal cortex, the left inferior frontal cortex [123] and especially the motor cortex [124]) but was elevated in the parietal lobe (the left precuneus and the right angular gyrus [123]) and the left frontoparietal network [123]. This agrees with the elevated sensorimotor connectivity of the left sensorimotor cortex with several regions of the right hemisphere (mainly the cingulate cortex, parahippocampal gyrus and cerebellum crus II) [125]. When compared with alterations of the CSTs (according to FA measurements), there were also changes in right sensorimotor connectivity; the connectivity was elevated if the CST was normal in a DTI analysis but was low if the CST was abnormal [125]. A network-by-network analysis revealed low activity in the default mode network [126] and elevated activity in the sensorimotor network (both of which include parts of the frontal lobe [126,127], although not the same ones). Thalamic connectivity is also elevated [127].
Resting-state activity (as measured by the ALFF) was elevated in the frontal lobe (the left anterior cingulate, right superior frontal cortex [31] left medial frontal gyrus and right inferior frontal gyrus [128]), the temporal lobe (the right parahippocampal cortex and the left inferior temporal cortex) and the occipital lobe (the middle occipital cortex) [31] but low in the right fusiform gyrus, the visual cortex and right postcentral gyrus [128].
The resting-state connectivity in the frontal cortex is correlated with disease progression and duration [128]. The connectivity between the left sensorimotor cortex and the right parahippocampal gyri and fourth cerebellar lobule is correlated with the ALSFRS-R [125]. Connectivity of the dorsal part of the precentral gyri was correlated with hand strength [124]. Activity in the right parahippocampal gyri was correlated with disease progression, whereas activity in the left anterior cingulate and left temporal gyrus was correlated with cognitive parameters [31].
Activation in fMRI In motor tasks, activation of the motor cortex was elevated [129][130][131] and spread widely across the supplementary motor area, the premotor cortex [132] and the parietal somatosensory cortex [129,132]. Inhibition was low [130]. Activation of the ipsilateral sensorimotor cortex was possibly linked to a compensatory mechanism in ALS [133]. A visual task was associated with poor activation of the secondary visual cortex and strong activation of associative areas, whereas an auditory task produced delayed activation in the secondary auditory cortex, and somatosensory stimulation produced prolonged activation of the right inferior frontal gyrus and right posterior insular cortex [134]. In a modified "go"/ "no-go" task, activation of the primary motor cortex, other motor areas, the frontal lobe, the insula and the cerebellum was found to decrease over a three-month period [135]. Using another type of stimulus, activation of the hippocampus also fell after 3 months [135]. In a theoryof-mind task designed to evaluation the mirror neuron system, ALS patients displayed abnormally strong activation of the right anterior cortical areas [136]. A link between the severity of motor weakness and the pattern of activation has never been observed, although the ALSFRS-R score was correlated with the signal change in the sensorimotor cortex in one study [129]. When considering the clinical phenotype, only a bulbar-onset subgroup showed low activation of the motor cortex during tongue movements [131]. Shifts in cerebellar or hippocampal activation were correlated with the ALSFRS-R score.

Discussion
When considering the data as a whole, the most diseasesensitive MRI patterns are located in motor regions (and especially along the CST from the cortex to the spinal cord). However, ALS also affects the brain more broadly; other parts of the frontal lobe, the temporal lobe, the hippocampus, the parietal lobe, the cingulum and the insula are commonly involved, whereas the occipital lobe is less frequently involved. For most MRI parameters, patient vs. control differences are mostly apparent in the left (dominant) hemisphere. The various MRI patterns appear to reflect disease severity, progression and duration. With fMRI, abnormally high levels of activation in the motor cortex are probably related to the diseaserelated loss of inhibition. Excessive activation appears to be associated with compensation for the loss of function caused by neuron loss.

Limitations of neuroimaging as a biomarker for ALS
The low average number of participants represents the main limitation: very few published studies have featured more than 30 subjects per group (i.e. ALS and control groups). Together with the differences in the proportions of the various patient phenotypes, this explains the high variability of the results -particularly outside the motor cortex and the CSTs [137].
Only a small proportion of morphometric studies included a whole-brain analysis. The proportion is higher for DTI studies and low (only one study) for MRS. The selection of regions of interest may over-emphasize changes in the CSTs and the motor system and underemphasize changes outside these areas.
Very few studies had a longitudinal design, which is nevertheless required for the full interpretation of abnormalities. The main difficulty in setting up longitudinal studies relates to the rapid disease progression and the appearance of respiratory failure, which prevents patients from lying in the MRI for any length of time. In studies longer than 6 months, the drop-out rate is usually over 50 %. fMRI studies have highlighted both elevated activity and low activity in several parts of the brain. The main physiopathological explanations relate to a physiological compensation response or elevated activity due to the loss of inhibitory mechanisms [127].
Spinal cord imaging is a rarely applied imaging modality in ALS, despite the pivotal involvement of the anterior horn of the spinal cord in ALS [137]. Spinal cord imaging appears to be most commonly used in studies of multiple sclerosis, since all studies in this disease context found differences (relative to controls) in the cervical spinal cord. To the best of our knowledge, no studies of the whole spinal cord have been publishedperhaps as a result of technical obstacles.

The biomarker concept
Currently, the diagnosis of ALS is based on the revised El Escorial criteria (also known as the Airlie House criteria), which were established in 1998 and published in 2000 [138]. Although diagnosis during the course of the disease has become relatively easy, the onset of ALS can be masked by clinical overlap with many other diseases [139]. The ideal biomarker is not only sensitive and specific (for various disease phenotypes) early in the disease process but must also easy to access [140]. In ALS, the typical time interval of 12 months between the initial symptoms and disease diagnosis makes this challenge tougher still [140][141][142]. The best way of determining early biomarkers for ALS may be to monitor cohorts of presymptomatic patients identified by genetic studies [143]. Although wet biomarkers based on blood or CSF analysis are relatively accessible, there are many candidates and study-to-study reproducibility appears to be quite low [5,140].

Perspectives
There is currently a need for a robust clinical-radiological description of ALS using MRI as a function of the disease progression, the various endophenotypes and left/right asymmetry (in terms of upper and lower motor neuron impairment, bulbar involvement, and the cognitive profile ranging from frontotemporal dementia, apathy and dysexecutive syndrome to mild attention disorders).
Despite the few number of studies, spinal cord could be a promising region of interest, although this approach is currently limited by technical post-processing difficulties and a sizeable number of large scale and longitudinal studies involving the spinal cord is needed to build stronger MRI biomarkers of the spinal cord. It may be of value to combine several techniques (e.g. atrophy mapping and measurements of the iron content in the grey matter). The results of MRI of the spinal cord are well correlated with the clinical impairment and might be a good surrogate marker.
A comprehensive analysis of MRI parameters in very large cohorts of ALS patients might reveal a radiological surrogate marker for use in clinical trials. Several points should be considered for large scale studies: in 2010, the first Neuroimaging Symposium in ALS (NiALS) had the aim of establishing consensus about the various applications of MRI to the study of ALS and the possibility of multicentre collaboration. Consensus criteria for the main MRI sequences (VBM, DTI, fMRI and spectroscopy) and the main clinical dataset were established [144]. The aim of this consensus was to lead large multicentre and longitudinal studies in imaging research or therapeutic trials [144]. <In our days, studies might better assess the most recent discovers in ALS such as genetic parameters (especially C9ORF72 expansion), the knowledge in cognitive and behavioural impairments even if the patients do not fulfil criteria for frontotemporal dementia [145]. There is still a need for longitudinal studies especially in presymptomatic and early phase of the disease [145]. Following this multicentre aim, a very recent study showed in 253 patients and 189 controls from eight international ALS specialised centre more widespread white matter tract change pooling the analyses than in single-centre analysis and reached estimation of neuropathological changes [146].

Conclusion
The MRI biomarkers appear to be well correlated with disease severity, duration and progression and are more sensitive in the brain and spinal cord motor regions. The built of MRI biomarkers is limited by the clinical various phenotypes but also by the lack of large and longitudinal studies. To date large cohorts with multicentre studies using a standardized MRI protocol are needed.