Knowledge of the layout and wiring of the somatosensory system is important to understand the post-stroke somatosensory deficits. The purpose of this section is to provide a comprehensive overview of somatosensory deficits, along with a brief overview of the anatomical and functional features of the somatosensory system.

The somatosensory system has two functions: somatosensory perception and guidance of action. The former means conscious perception and recognition, including exteroceptive and proprioceptive sensation (33). Exteroception sensation is the ability to recognize the superficial skin information from mechanoreceptors, thermoreceptors, and nociceptors, whereas proprioception sensation includes deep sensory input from muscle and joint receptors. Beyond perception, somatosensory processing also has the function of the guidance of action (34). Under normal conditions, the perception and action should act in a coordinated fashion to maintain the conduction of sensory information. Those functional dissociations suggest that separate neural pathways are involved in the processing of somatosensory input for conscious perception and the guidance of action.

Most information acquired from peripheral sensory receptors (except face) will be sent to the spinal cord, while unconscious proprioceptive information will be directly sent to the cerebellum through the spinocerebellar tract. From the spinal cord, the information will then ascend to the ventral posterior lateral (VPL) of the thalamus via the medial lemniscus or spinothalamic tract. The thalamus was long described as a relay station for sensory information transmitted from the periphery to the cortex, and it plays a critical role in nearly all aspects of cortical information processing, including carrying the information to the cortex and modulating the cortical processing (35–37). Thus, even a small thalamic lesion can cause deafferentation and a loss of somatosensory information (38, 39). The somatosensory areas of parietal cortex contain multiple topographic maps of the body surface, famously depicted as the “homunculus” of Penfield (40). Lesions of the parietal lobe are the most frequent cause of somatosensory deficits (35). Axons originating in the VPL will convey information mostly into the anterior parietal cortex (APC), which is viewed as the anatomical site of the primary somatosensory cortex (SI). It consists of four different areas (Brodmann areas 3a, 3b, 1, and 2) and each area contains a separate representation of the contralateral body (41).

Moreover, the secondary somatosensory cortex (SII), the posterior parietal cortex (PPC), and the insula are other key areas in the cortical somatosensory network and are thought to be involved in higher-order sensory processing. The SII is found in the parietal operculum and plays a role in information integration. It will receive the input mainly transmitted from the SI and directly receive some input from the thalamus (42, 43). The insula is located on the deep side of the Sylvian fissure and has many connections with multiple different brain regions. It plays an important role in the integration of somatosensory information with other sensory modalities, particularly the control of autonomic systems and pain encoding (44–47). The PPC is important for spatial processing and visually guided action, and the lesions of the PPC could lead to hemispatial neglect, with which patients will be unaware of sensory deficits (48, 49). Notably, evidence indicates the existence of at least two different streams in the processing of non-painful somatosensory stimuli. One stream projecting from the APC via the SII to the posterior insula is considered the ventral pathway and subserves perceptual recognition and learning (50–52). Another pathway terminating in the PPC is considered the dorsal pathway and subserves both perception and action (34). Both streams subserve the recognition and perception of somatosensory information, whereas action-related processing mainly occurs in the PPC, which can be assumed that the PPC could accumulate sensory input to encode action plans (53, 54). Besides, somatosensory processing also has a connection with the cerebellum (55–58). The cerebellum is associated with the somatosensory cortex afferently and efferently (59). It is responsible for monitoring sensory information and is also involved in forming predictive associations between sensory inputs and motor outputs (60, 61).

A limitation of structure MRI is that it might not necessarily reflect the full function of a certain brain area. This limitation has spurred interest in the use of fMRI, as a method to elucidate the nature of brain activity in relation to behavioral outcomes. Since its emergence in the early nineties, fMRI has become one of the most commonly used neuroimaging tools to map human brain functions (76). It could reflect the change of blood oxygenation in brain regions by detecting the blood-oxygen-level-dependent (BOLD) contrast. BOLD contrast forms the basis of fMRI formation, which can be divided into task-based and resting-state fMRI(rs-fMRI) (77). In the past two decades, task-based fMRI has been used widely to identify abnormal brain function and uncover brain substrates. However, task-based fMRI of stroke studies often suffers from a large variability because it is hard for stroke patients to perform the tasks well (78). Comparatively, rs-fMRI could recognize the spontaneous activities of the brain beyond instantaneous task-induced change, which is more suitable for stroke patients. Thus, in this section, the spotlight is mainly on the rs-fMRI studies. There are two major categories of rs-fMRI analytical methods: functional segregation and functional integration. The former focuses on the specific brain region activity, including Amplitude of Low Frequency Fluctuations (ALFF) and Regional Homogeneity (ReHo). Conversely, the measurement of functional integration can observe the brain as an intercomponent network, which relies on the analysis of different brain areas in connectivity (16).

The somatosensory maps can undergo cortical reorganization when sensory input patterns change, such as altered sensory use after peripheral or central lesions (118, 119). This process can be called neuroplasticity, which refers to the capability of the nervous system to reorganize and modify the brain structure and function, yet it still locks clear definition and general theory (120, 121). From a micro-perspective, the biological phenomena, such as long-term potentiation (LTP), dendritic remodeling, and neurogenesis, are strong evidence. In a macroscopic view, the MRI and other neuroimaging techniques can also prove that the brain can be reprogrammed and structurally rebuild by revealing the activity alteration of brain regions. The process of recovery and reorganization in response to behavioral demands or injury can be well-demonstrated by several advanced MRI approaches (16). For somatosensory map plasticity, the cellular and synaptic mechanisms of use-dependent map plasticity have been best described (122, 123). However, much less is known about neural mechanisms of central lesion-induced reorganization, particularly stroke-induced neuroplasticity. Only a few longitudinal studies have explored the time course of recovery from post-stroke somatosensory deficits. Those studies suggested that patients could recover at least partially after stroke, mainly during 3–6 months, but showed strong interindividual divergence (20, 124–126). Here, we want to discuss neuronal remodeling in response to spontaneous-induced recovery of somatosensory deficits by reviewing recent MRI studies. The elucidation of the recovery mechanisms of somatosensory deficits is important to provide bases for establishing proper rehabilitation strategies (127).

As already mentioned earlier in this review, previous studies have shown that thalamic lesions could lead to increased FC within the somatosensory-related cortex, which is presumed as the manifestation of compensatory effect (22, 25, 26). For example, He et al. (25) reported on contributions of peri-lesional reorganization. They found an association of somatosensory recovery with an increase in thalamic activation in the affected hemisphere in chronic thalamic infarction patients. It can thus be speculated that the ipsilesional thalamus itself may be involved in the neuroplasticity associated with somatosensory recovery. Consistently, a previous case demonstrated that ipsilesional thalamic reorganization could occur after a thalamic stroke and it may play a vital role in recovery from such a stroke (128). Also, several studies have emphasized the important role of ipsilesional thalamic in somatosensory recovery in patients with thalamic lesions (129–131). The thalamus is the central hub of sensory function. It acts not only as a simple relay transmitting information from the periphery to the cortex but also as a high-order relay among cortical regions (132). A previous MRI study suggested that infarcts causing pure sensory were predominantly in a thalamic site (133). Thus, the integrity of ipsilesional thalamic circuitry in the recovery of somatosensory deficits after thalamic lesions appears to be important. Moreover, Chen et al. (22) attributed the sensory recovery to other unaffected somatosensory cortex or pathways. They found that in stroke patients with lesions in or inferior to the thalamus, increased FC could be observed between the lesioned locations and other remote areas, including the contralesional middle temporal gyrus and superior occipital gyrus, which could be interpreted by the concept of the cortical-thalamo-cortical loop. According to the cortical-thalamo-cortical loop theory, the thalamus is connected anatomically to the somatosensory-related cortex via thalamo-cortical projections (134–136). The temporal gyrus and the occipital gyrus are considered important brain regions in the thalamo-cortical circuit (89). Thus, it is reasonable to assume that the likely relevance of somatosensory recovery and alterations in peri-lesional thalamic and some remote regions might reflect the involvement of thalamo-cortical circuit in reorganization patterns induced by subcortical lesions. In addition, Lee et al. (27) also attempted to explore the underlying mechanisms of somatosensory recovery, using fMRI in 11 chronic thalamic hemorrhages. They performed fMRI by proprioceptive input and found that the recovery of proprioceptive function mainly depends on the enhancements of contralateral primary sensorimotor cortex (SM1) activation. It should be noted that proprioceptive information has a lot of effects on coordinated motor function and the degree of proprioceptive impairment can be regarded as an important predictor for motor recovery (137–140). The correlation between the improvement of somatosensory function and the activations of SM1 could lead to the conception of sensori-motor interaction. Evidence indicates that a loss of somatosensory information caused by sensori-motor disconnection could worsen motor impairments (11). Therefore, it can be assumed that the activation of SM1 during the somatosensory recovery may partly facilitate motor training effects on motor function (141, 142).

The aforementioned studies provide a general hypothesis that the subcortical-cortical circuit and the sensori-motor networks may account for somatosensory recovery after subcortical lesions. These observations have aroused scientific interest in investigating the multiple patterns of brain activation across stroke patients with different lesion locations. Evidence from the model of motor recovery showed that different patterns of activation were associated with different lesion locations (143). It, therefore, stands to reason that lesions following subcortical or cortical regions may have different activation patterns during recovery of somatosensory. Carey et al. (28) conducted a fMRI study to characterize and compare touch impairment-related brain activation between 11 stroke patients (~1-month post-stroke) with subcortical lesions and 8 with cortical lesions. The results showed that in the subcortical group, touch discrimination correlated negatively with the activation in the ipsilesional SI, SII, contralesional thalamus, and frontal attention regions; but in the cortical group, there was no significant correlated activity. However, this finding had some subtle differences from their later study (29). Five years later, they conducted a similar fMRI study to compare four stroke patients with subcortical lesions (thalamic/capsular) and seven with cortical lesions (the SI/SII) in the chronic phase of recovery and reported activation of ipsilesional supramarginal gyrus in stroke patients with thalamic or capsular lesions and no common activation in stroke patients with SI or SII lesions (29). Despite the small sample size, those studies provided novel insight into lesion-specific mechanisms of brain reorganization and potential therapeutic strategies to target specific brain regions, which warrant further investigation.

In addition, some imaging studies on the subject of recovery have compared stroke patients at an arbitrarily chosen time point (mainly at the chronic stage) with healthy controls to investigate the changes in neural activity and functional connectivity. However, those results were not sufficient to account for the possible time-dependent dynamics during the recovery, which called for more studies that would longitudinally investigate stroke patients in the process of recovery. For this purpose, Bannister et al. (32) conducted a longitudinal fMRI study to identify long-term changes in FC of the somatosensory network in 10 stroke patients with impaired touch sensation at 1- and 3-months by using seed-based analysis and touch sensation was evaluated by TDT. Compared to HC, 1-month post-stroke patients exhibited disruption of inter-hemispheric FC of homologous SI regions, whereas, at 6 months, there was some return of interhemispheric SI connectivity. This “return to more normal patterns” may be viewed as being associated with better recovery in the chronic stage (144, 145). Furthermore, correlation analysis was performed and great improvement in TDT scores was reported to be associated with significant FC changes within the contralesional hemisphere, i.e., great FC between contralesional SI and cerebellum at 1 month and great FC between contralesional thalamus and SII at 6 months. Those findings highlighted the important role of the contralesional hemisphere in post-stroke somatosensory performance and recovery, which was also purposed in the previous case reports (146, 147). Besides functional plasticity, there is a hypothesis that changes in brain morphology over time may also impact recovery. As we mentioned above, a longitudinal study conducted by Kessner et al. demonstrated different lesion-symptom patterns over time. They recruited 101 patients at the baseline, and finally, only 46 patients completed the entire study with all 3-time point measurements (at first 5 days and 3 months, 12 months follow-up) and used RASP to measure sensory deficits. Using VLSM, they found that a cluster of brain regions was significantly associated with lower RASP scores at baseline and 3-month post-stroke but no clear association between stroke lesion and sensory performances were observed at 12 months, which suggested that other factors beyond lesioned locations may account for somatosensory recovery. The brain atrophy rate from 3 to 12 months post-stroke was reported to be more than twice that of age-matched HC (148). There is some evidence that areas, such as the thalamus, hippocampus, and insula show greater atrophy after stroke (149–151). Yassi et al. (31) explored the long-term changes in the contralesional hemisphere in 15 acute middle cerebral artery territory ischemic stroke patients by observing volume change and FC within 1 week of onset and at 1 and 3 months. They have identified significant contralesional thalamic atrophy in the first 3 months and a positive correlation of atrophy degree with initial stroke severity. This result corroborated the finding that hypoperfusion in the thalamus could be observed after stroke in the acute and subacute stages, which accounts for a good or bad recovery (152–156). However, a recent retrospective cross-section study didn't identify a significant longitudinal relationship between total thalamic volume and time since a stroke in the chronic stages (157). This result was unexpected to some extent as previous studies have reported a negative correlation in the acute and subacute stages. All those findings raise the following possibility: structural thalamic adaption may be restricted to the acute or subacute stages and the changes in chronic stages (particularly after 3 months) may be too subtle to be detected using current techniques (157). In additional, Longitudinal changes in structural and functional features in stroke patients with somatosensory deficits were shown in Supplementary Table S1.

Stroke is the most common cause of various degrees of functional impairments and neurological deficits, ranging from mild to severe. Somatosensory deficits after stroke are a common symptom caused by impairments of the central nervous systems. Previous studies utilizing neuroimaging tools have revealed detailed knowledge about a network of brain areas involved in the processing of somatosensory information. As mentioned in the first section, the whole process involves a functionally diverse set of brain regions, including the thalamus, SI, SII, PPC, insula, and cerebellum. Although there is a good description of a brain network implicated in the processing of somatosensory information, we still know less about the underlying mechanisms of somatosensory deficits and map plasticity during recovery. To summarize, this review has provided an up-to-date summary of the available data from MRI studies. Both structural MRI and fMRI studies have generated achievements in fundamental neuroscience research and clinical applications.

For structural MRI, the clinical applications of VLSM among stroke survivors have provided an important method for exploring potential structural abnormalities of neurologic somatosensory disorders. These studies generally suggest that stroke lesions occurring in the SI, SII, and IC could lead to somatosensory symptoms, but the one-to-one correspondence between a specific lesion pattern of brain regions and a certain somatosensory modality remains unclear. Moreover, the application of the structural brain connectome suggested that network disruption could explain the remote effects of a stroke lesion on additional variation in clinical outcomes. Recently, there is an apparent shift in the focus of fMRI. A series of ReHo and ALFF studies agreed on the notion that a positive correlation between somatosensory functions and cortical activation in stroke patients. Equally important was the finding that somatosensory deficits are a common symptom following lesions of the parietal lobe. Then, we also reviewed the state-of-the-art methods developed to analyze FC by using fMRI. A reduction of inter-hemispheric symmetry across the somatosensory network governing sensory perception could be observed in post-stroke somatosensory deficits. Some other functional imaging studies have revealed an increased FC of sensory-related regions in chronic stroke patients with sensory deficits.

Those results also emphasized the need to view the involvement of the neuroplasticity in somatosensory recovery after stroke, although the specific concept of the compensatory effect remains speculative. Current knowledge of post-stroke motor recovery and different patterns of neuroplasticity has been extensively identified (158, 159). However, little is known about sensory recovery. Limited studies have suggested that the brain could reorganize to optimize somatosensory recovery and the process may be influenced by the lesion location and course of stroke (15, 28, 160, 161). Thus, recovery varies among individuals, resulting in the heterogeneity of stroke outcomes. It is acknowledged that gray matter atrophy co-exists with brain plasticity presenting with structural and functional remolding in specific regions (162). The substantial recovery of subcortical lesions, particularly thalamic stroke, maybe due to the involvement of the thalamo-cortical circuit and sensori-motor networks. In addition, activation patterns of corresponding brain areas may differ between cortical lesions and subcortical lesions. We believe that a better understanding of the underlying neurological mechanisms could contribute to the development of novel therapeutic strategies that will promote repair and alleviate somatosensory deficits.

Although studies obtained by this review provide potential insights into the neural substrate of post-stroke somatosensory deficits, there are many conflicting results obtained by extensive studies. Those controversies may probably be attributed to the study design, suggesting current research has some shortcomings that are yet to be solved. Therefore, several limitations and challenges will be listed in the following paragraphs.

In contrast to a large number of studies that applied those approaches in the field of motor symptoms and motor recovery after stroke, a relatively small number of studies have investigated somatosensory deficits after stroke. There are several possibilities to interpret this apparent situation. One important reason is that the clinical need for motor recovery is greater than that for sensory recovery, though intact sensory function could promote motor recovery (163). Another possibility is the difficulties in recruiting patients with pure sensory stroke. The presence of somatosensory deficits, combined with motor dysfunction, makes interpretation difficult. The second drawback of this review is the small sample size of original MRI studies in this field of research. Many included studies have a limitation of a small number of patients because the costs of MRI were prohibitive for larger studies. Recent evidence indicates that limited sample sizes may yield mixed results in MRI studies, although there is no consensus on the sample size calculation (164). The small samples may lack enough statistical power. Thus, it is not sufficient to draw firm conclusions as the evidence is mixed.

Beyond the reasons discussed above, we speculate that different results obtained by different studies may also depend on the heterogeneity in clinical measurement of somatosensory deficits, which is the most complicated part of the neurologic investigation. Compared to assessments for motor dysfunction, somatosensory assessments in current clinical practice are less reliable and reproducible due to the lack of consensus on sensory paradigms (165). Obviously, different studies adopted different scales to measure somatosensory performances, resulting in different results. Thus, to reduce the clinical heterogeneity of somatosensory outcomes, it's imperative to establish consensus on the scales for scientific research. Furthermore, the complexities of cortical interactions and the multitude of sensory modalities also require doctors or therapists, who are equipped with a great amount of knowledge and experience in neurology to make a diagnosis. Equally important is the cooperation and compliance of the patients and they should try to understand that this procedure is the least objective part of a neurologic examination. Taken together, the complexity of sensory assessments may contribute to the paucity of quality research in this field.

There is considerable variability among the currently available findings. Those discrepancies may be attributable to the small sample size of studies, heterogeneity in the clinical phenotypes, and differences in the clinical assessments or analytic methods. The assumptions formulated by included studies can also provide several suggestions for future directions of the research in this field.

Firstly, it should be noted that the thalamus didn't show up in most VLSM studies, despite its importance in somatosensory processing. This could be possibly due to the small number of patients who had lesions in the thalamus in those studies. Thus, future research should address the MRI scanning and somatosensory assessment with respect to lesion-symptom associations in patients with thalamic stroke. Besides, whether a specific lesion pattern may account for a certain somatosensory impairment remains elusive. More studies using VLSM, or other analytical methods should focus on the one-to-one correspondence between a specific lesion pattern of brain regions and a certain somatosensory modality. We think this may help to explain the variability in deficits across patients. And then, many FC studies have suggested that a focal stroke lesion affects not only the peri-lesional region but also the pathway or network to which it belongs. For example, the lesions in the thalamo-cortical circuit were considered vulnerable to cause somatosensory deficits, and increased FC within this circuit could be observed in the process of somatosensory recovery (19, 27). It is acknowledged that the thalamus and some cortical regions (e.g., the temporal gyrus and the occipital gyrus) could share and communicate sensory information and are linked on an anatomic basis. There is a hypothesis that somatosensory deficits can be regained due to the compensation of the thalamo-cortical circuit. But this hypothesis requires further experiments to confirm. Similarly, one previous fMRI study emphasized the importance of the sensori-motor network in somatosensory recovery (27). Accumulating evidence suggested that intact sensory information could enhance motor cortex excitability. Although the exact mechanisms of sensori-motor interactions remain unknown, there is increasing evidence that sensori-motor disconnection could cause sensory or motor dysfunction (11, 142). More future work should focus on the interaction between sensory and motor systems: how does a simple loss of one system cause functional impairments in the other one; how does sensori-motor execute its function in the process of somatosensory recovery. Moreover, we speculate that the stroke lesions following subcortical or cortical regions may have different activation patterns during recovery of somatosensory and we think it is worthy of further examination to address this issue. This could be executed through more restrictive inclusion criteria in future research, for example, the lesion location should be restricted to a specific subcortical area. More restrictive inclusion criteria could lead to more homogenous groups or rigorous experimental control resulting in higher sensitivity. Finally, all those speculations need to be confirmed by more high-qualified studies with larger sample sizes, especially longitudinal studies. It is acknowledged that longitudinal studies using MRI techniques could assess structural and functional changes in the brain over time in order to investigate the “return to normal pattern.” Thus, longitudinal studies with a large sample size that span the acute to chronic phases are required to determine the underlying possible compensatory mechanisms of somatosensory deficits and represent a future target for therapy.

Whilst limited studies hamper our understanding of the pathogenesis of post-stroke somatosensory deficits, some methodological issues in the existing MRI studies could contribute to the understanding of the progress in the application of MRI techniques in this field. In this paper, we summarized an updated account of advances in MRI. We hope that this review will equip new researchers with a more complete understanding of advanced MRI techniques and assist them in launching more comprehensive studies. As seen from the studies presented here, many studies only used one single MRI modality, making it difficult to directly investigate the synchronous structural and functional changes in the same samples. Considering that different MRI modalities provide different aspects of neural alterations, it is anticipated that methods combining functional and structural data to find out the neuroimaging biomarkers will be of value. With regard to fMRI, the number of studies that applied the ReHo and ALFF was relatively small. We hope that those two methods will be used far more widely for analyzing the activations of specific brain regions and revealing the functional segregation. In comparison, FC has contributed more to the field of post-stroke somatosensory deficits. Also, how to choose an appropriate method to analyze FC has been a long-time discussion for novice researchers (76, 166, 167). The preference and selection of a method are dependent on the scenario and study design. As we mentioned previously, most studies used seed-based methods, whilst only one study has tried to apply the graph theory to examine the whole brain. Undoubtedly, selections of seeds in different studies were identical, which may add to selection bias, and then the results may be different. Graph theory can help in modeling the brain as a complex network represented graphically, thereby solving the problem caused by seed-based analysis. Thus, there is a great need for more research using the graph theory to improve our understanding at the whole-brain level. Furthermore, the hypothesis on lateralized processing of sensation is also worth noting. We have noticed that many studies didn't take hemisphere-specific information into consideration. In many conditions, if the lesion locations were disturbed in both hemispheres, data would be flipped to one side so lesions would be represented within a single hemisphere (19, 20, 22, 26, 28, 29, 32). This “flipping” method seems to make it simpler and more convenient to analyze the data, but the differences in FC associated with the side of the lesion were therefore hidden (24). Despite the effects of flipping have not reached a consensus, we still think it is better not to flip the MRI data before pre-processing, since patients with right hemisphere stroke may be more likely to have somatosensory deficits (19). Therefore, future studies using a multimodal MRI approach, unbiased whole-brain analytical methods, and taking into account the lateralization when examining somatosensory deficits will be required to elucidate the exact mechanisms.

Although MRI is considered a relatively mature method, ongoing improvements in analytical strategies and hardware continue to improve our knowledge of the mechanisms underpinning brain function and disease (100). Other well-developed methodological approaches, such as TMS, diffusion tensor imaging (DTI), Arterial Spin Labeling (ASL), magnetic sensitivity weighted imaging (SWI), quantitative sensitivity magnetization (QSM), and magnetic resonance spectroscopy (MRS) should be given due importance and incorporated into future studies. Besides analytic methods referred to above, other methods can also be applied to analyze the MRI data. For example, another two methods to analyze rs-fMRI data are known as clustering algorithms and multivariate pattern classification (168). The former could group a collection of voxels on the basis of similarities, and the latter could be used to predict individual brain maturity. Moreover, with the rapid advancement of artificial intelligence, machine learning, and especially deep learning technologies are more and more widely used in MRI studies (169–174). Such new and valuable tools may spur radioligand development for extending our knowledge of the pathogenesis of somatosensory deficits and informing treatment strategies.