Ever since the definition of RIS in 2009 by Okuda et al. (1), researchers have tried to characterise and anticipate its clinical evolution. As MRI began to play a crucial role in predicting MS disease activity and progression from its onset (6, 7), many researchers investigated which MRI hallmarks distinguish individuals with RIS from healthy controls and people with MS. Whilst early research focussed on lesion characteristics (2, 8), later studies applied quantitative multi-modal MRI techniques to study normal-appearing tissues. The methods used included magnetisation transfer (MT), diffusion-weighted imaging (DWI), magnetic resonance spectroscopic imaging (¹H-MRSI), and brain and spinal cord volumetrics.

An early study by De Stefano et al. (9) at 1.5T MRI reported that people with RIS and relapsing-remitting MS (RRMS) had similar global brain and cortical volumes, both lower than healthy controls. However, people with RRMS had a lower MT ratio (i.e., an index reflecting damage to the myelinated fibres) (10) in the normal-appearing brain tissue compared with RIS and healthy control participants. The authors concluded that this index of magnetisation exchange between brain macromolecules and water could be used to stratify the risk of MS conversion among RIS patients. However, their study had a cross-sectional design, with findings not yet confirmed in a longitudinal setting.

The same group further explored alterations in the normal-appearing brain tissue of individuals with RIS using ¹H-MRSI at 1.5 T (11). N-acetyl aspartate (NAA) and choline (Cho) normalised to creatine (Cr) were measured in lesional/perilesional, normal-appearing white matter (WM) regions and the cortical grey matter (GM). They found lower NAA/Cr levels in RIS than in controls. NAA is synthesised in neurons and is considered a marker of neuronal integrity in the central nervous system (12). Its alterations have been widely reported in MS (13). Stromillo et al. (11) stratified RIS subjects according to the risk of developing MS based on established prognostic factors: spinal cord lesions, unmatched oligoclonal bands, and DIT. However, they could not find any differences in patients with different risk levels.

Another study (14) at 1.5 T confirmed the volumetric findings from De Stefano et al. (9). However, whilst the latter had included RRMS patients with a disease duration of up to 7 years, Rojas et al. (14) compared RIS subjects with people with clinically isolated syndrome (CIS) within 2 months of onset. They found similar global brain and cortical volumes for RIS and CIS, both lower than healthy controls.

The studies mentioned above did not investigate deep grey matter. However, Azevedo et al. (15), using 3T MRI (see Table 1), found a weak effect towards lower total GM volume in RIS compared with healthy controls. However, deep GM volume was significantly lower in RIS than in healthy controls, especially the thalamus. The authors also reported thinner right superior and inferior parietal cortical gyri in RIS than in controls.

Labiano-Fontcuberta et al. (17) compared RIS with CIS at 12 months after onset with 3T MRI. They reported no differences in brain T2-white matter (WM) lesion load, but T2-lesion load correlated with cortical volume in both groups. Whilst participants with RIS had lower cortical and thalamic volumes than controls, those with CIS did not. Specific cortical thinning patterns also emerged in the RIS population but did not survive Bonferroni correction for multiple comparisons. Although WM volumes were similar between groups, voxel-based analysis using diffusion-weighted imaging (DWI) showed lower fractional anisotropy (FA; i.e., a marker of WM fibre integrity) (31) in the cerebellar WM in the CIS population compared with healthy controls.

The same group also characterised differences between RIS and healthy controls using ¹H-MRSI and diffusion tensor imaging (DTI)-derived FA and mean diffusivity (MD) maps with 3T MRI (20). People with RIS showed lower brain and cortical volumes than controls, but contrary to previous findings by Stromillo et al. (11), they did not show alterations in NAA/Cho and Cr/Cho nor in white matter FA and MD.

A recent study (21) applied a machine-learning approach to differentiate RIS from CIS using multi-modal 3T MRI, which included cortical thickness, cortical and subcortical GM volumes, and DTI measures of WM integrity. A model of three features (left rostral middle frontal gyrus volume and FA of the right amygdala and lingual gyrus) could discriminate between RIS and CIS with an accuracy of 78%.

Interestingly, another recent study (19) exploring brain volumetrics at 3T in RIS reported a decrease in cerebellar WM and anterior GM compared with healthy controls. No differences were observed in total brain volume and cortical or deep GM volume between RIS and healthy controls.

Spinal cord lesions in RIS increase the risk of conversion to MS (2). In total, two studies have also investigated spinal cord atrophy and quantitative spinal cord MRI alterations in RIS.

Zeydan et al. (22) did not find differences in spinal cord average cross-sectional area at C2 and C7 between RIS and RRMS even though people with RRMS had more frequent cervical cord lesions (91%) than those with RIS (44%). Cervical spinal cord atrophy was only present in people with secondary progressive MS.

Alcaide-Leon et al. (23) assessed spinal cord DTI and MT metrics in RIS vs. healthy controls. The spinal cord cross-sectional area was measured between C3 and C4 and reported decreased brain volume in RIS compared with controls, with no evidence of cervical spinal cord atrophy. Furthermore, quantitative MRI analysis revealed weak evidence for lower spinal cord MT ratio in RIS compared with healthy controls.

Iron-sensitive MRI sequences (such as T2^*- and susceptibility-weighted imaging) can identify veins within demyelinating white-matter (WM) lesions, called the “central vein sign” (CVS) (32). The proportion of WM lesions that display the CVS can help distinguish between MS and other conditions associated with white-matter disease. A cutoff threshold of 40% of WM lesions displaying the CVS sign has been proposed (32)—proportions above this would be consistent with MS. An alternative proposal suggested a minimum of six lesions showing the CVS to support MS (33).

In total, three studies have considered the CVS in RIS. Suthiphosuwan et al. (24) analysed the number of WM lesions displaying the CVS in RIS and compared this with the reported proportion of CVS-positive WM lesions in RRMS. Out of 20 participants recruited, 18 (90%) had at least 40% of WM lesions showing the CVS and 19 (95%) had at least six lesions with the CVS. The two people who did not meet the 40% threshold had a relative paucity of imaging features consistent with RIS—for example, most lesions were small and punctate and were located in the anterior subcortical and deep WM. Regardless, these two participants had 29 and 31% of lesions demonstrating the CVS.

The same group also assessed the association of the CVS with cognitive impairment (25). In their cohort of 27 people with RIS, 25 (97%) had at least 40% of WM lesions displaying the CVS. The proportion of CVS-positive WM lesions predicted performance on the California Verbal Learning Test, an assessment of verbal memory.

The third study (26) performed a retrospective analysis of people diagnosed with RIS at their centre. Most of the 89 people with RIS did not undergo imaging sequences that would identify the CVS. Of the five that had T2^*-weighted imaging, all met the same 40% threshold of WM lesions with CVS.

Advanced MRI techniques can also be used to assess brain connectivity. This might be with multi-shell DWI to estimate trajectories of WM tracts and examine structural connectivity or with resting-state functional MRI to review functional connectivity between different brain regions (34). Graph theory can then be applied to interrogate the integrity of the whole brain network (35)—for example, a comparison of CIS brain networks and those derived from healthy controls identified differences in network organisation between the two groups (36).

Only one study to date has assessed brain connectivity in RIS. Giorgio et al. (27) compared connectivity between RIS, RRMS, and healthy controls. Their voxel-wise analysis identified no differences between people with RIS and RRMS, though differences were noted with analysis of WM tract integrity. DTI metrics [FA, MD and radial diffusivity (RD)] were altered in RIS in regions with WM lesions but preserved in normal-appearing WM; they were altered in both WM lesions and normal-appearing WM in people with RRMS. Furthermore, a less conservative statistical analysis (significance p < 0.05, rather than < 0.01) showed lower axial diffusivity (AD) and RD in certain WM tracts incorporating both WM lesions and normal-appearing WM, such as the corpus callosum and corticospinal tract, in RIS compared with RRMS.

The same study (27) also examined functional connectivity using resting-state functional MRI. In total, two of the 12 resting-state networks had lower functional connectivity in RIS than RRMS: the sensorimotor network and right working memory network. However, there were no differences in these networks between RIS and HCs. There were no voxel-wise differences in brain functional connectivity between HCs and people with RIS.

Retinal layer thinning, as measured by optical coherence tomography (OCT), is considered a biomarker for neurodegeneration in MS (37). Some studies have assessed its role in RIS. Filippatou et al. (28) assessed OCT in RIS and healthy controls. Interestingly, no OCT differences were seen between RIS and healthy controls. However, there were associations between the presence of either spinal cord or infratentorial lesions and a reduced macular ganglion cell-inner plexiform layer (GCIPL). As spinal cord lesions increase the risk of conversion from RIS to MS (2), GCIPL thinning may similarly indicate high risk, though there was no follow-up to assess this. No associations were seen between OCT metrics and other risk factors for conversion to MS, such as CSF-unique oligoclonal bands or MRI-enhancing lesions.

Knier et al. (29) analysed the association of retinal thicknesses with inflammatory activity in RIS and CIS. They noted macular retinal nerve fibre layer (mRNFL) thinning and inner nuclear layer (INL) thickening in RIS and CIS compared with healthy controls. GCIPL was the lowest in the RIS group, though it was lower in CIS than HC. When MRI was repeated after 1 year, the appearance of new inflammatory lesions correlated with a thinner baseline GCIPL and a thicker baseline INL. No clinical activity was reported in the RIS group—the MRI changes merely represented DIT rather than conversion to MS; however, DIT is also a risk factor for conversion from RIS to MS (1).

Overall, two more longitudinal studies have been reported. Vural et al. (18) found reduced brain and thalamic volumes in RIS compared with healthy controls. OCT analysis revealed thinner GCIPL, mRNFL, and temporal peripapillary RNFL (pRNFL) in RIS. In RIS, these retinal metrics correlated with brain and thalamic volumes. Participants with RIS were followed-up for a median of 5 years: four out of 15 participants experienced a clinical episode, converting to MS. The pRNFL was thinner in individuals with RIS converting to MS than in those who did not convert. The groups were otherwise similar regarding T2-lesion load, with a presence or absence of gadolinium-enhancing lesions and volumetric measurements at the baseline.

Another longitudinal study (30) followed up participants with RIS for 6 years. People with RIS had thinner pRNFL and GCIPL than healthy controls. A total of eight out of 36 individuals converted to MS at follow-up. Conversion to MS was associated with thinning of the pRNFL and GCIPL at baseline and over time. After adjusting for other factors (age, sex, immunotherapy, and the occurrence of spinal cord lesions), Cox proportional hazards regression revealed a hazard ratio of 1.08 for conversion to MS for each 1 μm decline in pRNFL.

Figure 1 summarises the mini-review findings.