Peripheral Neuropathy in Diabetes: What Can MRI Do?

Diabetic peripheral neuropathy (DPN) is by far the most common acquired neuropathy related to diabetes (1), and the severity of DPN is significantly associated with the prognosis of patients with diabetes (2). Therefore, the early and accurate diagnosis of DPN is critical to identify high-risk groups and prevent amputation and foot ulcers among those at risk.

Normally, the diagnosis of DPN is based on neurological examinations, nerve conduction studies (NCS), quantitative sensory test (QST), a clinical neurological function scoring system, corneal confocal microscopy (CCM), and intraepidermal nerve fiber density (IENFD). These diagnostic methods are useful tools for diagnosing and evaluating the progress of neuropathy in patients with diabetes, but there are still many defects. For instance, it has recently been proven that the neurological deficit assessment scale has poor diagnostic repeatability. NCS is the current gold standard for identifying DPN (3). However, there are a few drawbacks: it is time-consuming, expensive, and requires operation by professionals. NCS cannot exclude early DPN because it is not sensitive to small-fiber neuropathy (4). QST is used to evaluate sensory neuropathy. The repeatability of QST in clinical practice has always been problematic, as its treatment courses for the same patient can differ significantly. It has no value in disease localization (5). The clinical neurological function scoring system is used primarily in epidemiological surveys to screen, evaluate, and grade DPN severity. The standardizing of clinical assessment with scored assessments remains subjective and is highly dependent on the examiners' interpretations (6–8). CCM is a rapid, noninvasive nerve imaging technique for assessing corneal nerve fiber morphology (9). In addition to its noninvasiveness, ease of operation, and short surface anesthesia time, CCM provides an accurate and dynamic view of corneal nerve changes over time. IENFD is commonly used as a surrogate measure for small-fiber injury in natural history studies and clinical trials. Invasiveness is the primary problem for IENFD as a biomarker for small-fiber neuropathy. This limits its practical application, especially when repeated biopsies are required in longitudinal studies or clinical intervention trials (10).

Therefore, we need a noninvasive, easily accepted, accurate, and intuitive evaluation method, and magnetic resonance may meet our requirements. This article discusses the diagnosis and follow-up of DPN with different MRI sequences, with emphasis on lower-limb nerves, and we also discuss the application of MRI to the central nervous system in patients with DPN (Fig. 1).

In the early 1990s, Filler et al. (11) established the term magnetic resonance neurography (MRN) to mean sequences with higher structural screen resolution and higher neuropathy contrast. MRI sequences used in DPN research mainly include high-resolution MRN and quantitative MRI (such as quantitative relaxometry, diffusion-weighted imaging [DWI], diffusion tensor imaging [DTI], etc.), which extends the role of MRN in DPN. They exclude suspected mass lesions that cause nerve compression and can be used for early screening, severity grading, and follow-up of DPN.

MRN examinations with high-resolution and high-contrast sequences, including sequences that are heavily weighted by relaxation time T2, with fat saturation are excellent techniques for showing peripheral nerve anatomy (12), and they are also the most mature and commonly used techniques in MRN. The injury of peripheral nerve is mainly judged by observing the changes in relaxation time T2–weighted imaging (T2WI) signal and nerve morphology. For most traumatic and focal neuropathies, the interpretation and morphometric criteria of intraneural T2WI signal help establish diagnosis and guide treatment. Although T2WI signal is highly sensitive for detecting nerve injury, its specificity is low and cannot be quantified. As a result, evaluation remains qualitative and subjective, and comparing differences across different scanner hardware and acquisition sequences is difficult and frequently limited to descriptions of relative contrast among different regions of the body. Measuring neuromorphic changes in diabetic neuropathy also facilitates the diagnosis of DPN but with low sensitivity (13).

In MRI, relaxometry is a common measurement. The definition of quantitative MRI is the process of decoupling the different contrast mechanisms that constitute the overall MRI signal so that biophysical parameters can be measured (14). Several fundamental quantitative MRI parameters reflect the local tissue environment. These include relaxation times T1, T2, and T2* and, by extension, proton spin density (PSD). Water content, dissolved oxygen, and macromolecular content affect the relaxation time T1. Water content, iron levels, and tissue composition and structure affect the relaxation time T2. Water content, deoxyhemoglobin, and iron levels affect the relaxation time T2*, and it is even more sensitive to iron levels than T2 (14). Another tissue-intrinsic parameter is PSD, which refers to the concentration of protons excitable by MRI. There is no effect of transverse relaxation on it, and it is considered to reflect total water content, including protons bound to macromolecules such as myelin (15,16). Measurement of the relaxation time often requires the acquisition of a series of images, with a slightly different weighting for each image, to sample the recovery or decay of the MRI signal. In this series, many images are often required to ensure adequate sampling of the signal evolution. The PSD is considered a semiquantitative parameter. It depends directly on the MRI signal and related parameters, so it is difficult to standardize (17).

In the 1990s, Basser et al. (18) introduced DTI, a quantitative MRI technique that measures both the direction and magnitude of proton diffusion using magnetic gradients in multiple orientations. Mori et al. (19) likened it to the shape of ink dropped on paper. We can understand the isotropic diffusion and the anisotropic diffusion through the shape of the ink trace, and we can judge the fiber structure below the paper. The peripheral nerves are like the fiber structure below the paper. We need to understand nerve fibers indirectly by understanding the diffusion of water. In DTI technology, diffusion gradients are applied in six or more directions, and three eigenvalues are generated in the perpendicular direction based on tensor calculus. Based on these eigenvalues, diffusion coefficient (MD), fractional anisotropy (FA), axial diffusion coefficient (AD), and radial diffusion coefficient (RD) can be calculated (20). The diffusion of nerve fluid is highly anisotropic according to physiology. There is a strong diffusivity in the fiber direction (AD), but orthogonal diffusivity in the direction of the fiber is hampered by numerous parallel cell membranes (RD). The FA is the most frequently used DTI parameter and is considered a general biomarker of nerve tissue integrity (19). MD is the mean of the three diagonal elements of the diffusion tensor (equivalent to apparent diffusion coefficient [ADC]). It is possible to evaluate the integrity of neural tissue more accurately by quantifying the direction of parallel nerve fibers and the direction of vertical nerve fibers (21). However, there is still no consensus on the selection of b value, which affects the accuracy and reliability of quantitative diffusion parameters.

Usually, nerve biopsy is not indicated in diabetic neuropathy, and in length-dependent diabetic polyneuropathy (LDDP), nerve biopsy is generally acknowledged as useless (22). Therefore, there are few studies on the MRN sequence and the pathology of the lesions in DPN patients. At present, relevant research mainly focuses on animal experiments. Wang et al. (23) carried out sequential MRI examination and histological assessment on streptozotocin-induced diabetic rats. They found that T2 value gradually increased 2 weeks after sciatic nerve induction in diabetic rats. They inferred that the increase in T2 value in early DPN was due to endoneural edema, and in late DPN the increase was due to progressive neurodegeneration and insignificant coexistence and regeneration. Schwarz et al. (24) also performed MRI examination and histological assessment on streptozotocin-induced diabetic rats, but they got different results. In the experimental mice, the T2 time of the sciatic nerve decreased significantly, and the signal intensity of T2WI decreased, while other MRI parameters, such as proton density, did not change. Histologically, they believed this was due to an increase in the surface area of the cell membrane within the nerve resulting from a decrease in the radius of nerve fibers, an increase in the number of nerve fibers, and a decrease in the average myelination. The test conditions and equipment of the two studies were different, so they were not comparable. Both of them showed the relationship between T2 value and DPN histology.

High-resolution MRN has been proven effective in detecting and localizing peripheral nerve lesions in DPN (25). The use of a high-field clinical scanner (3 T) and T2WI sequence in peripheral neuroimaging can show an increased MRI signal in a variety of focal and nonfocal neuropathies (26). In a T2WI fat–suppressed sequence, the signal intensity of vital nerves and muscle is similar (27). Jende et al. (28) defined T2WI-hyperintense lesions as nerve bundles with high T2WI signal intensity, at least 25% higher than that of muscle tissue, and T2WI-hypointense lesions were defined as nerve bundles with T2WI signal intensity at least 25% lower than that of muscle tissue. Neuropathy with high signal intensity in T2WI has been shown to be negatively correlated with nerve conduction parameters, and the number of lesions and extension length are directly proportional to the severity of the lesion (29).

In DPN patients, peripheral nerve MRI signals change, and these changes are usually accompanied by morphological changes. Generally, in type 1 diabetes (26) and type 2 diabetes (13), the cross-sectional area (CSA) of DPN patients was larger than that of healthy controls. This manifestation usually occurs in patients with DPN in the acute or subacute stage, while in the chronic stage, it is characterized by atrophic fascicles with intraepineurial fat deposits (12). Changes in the CSA of the peripheral nerves were related to the nerve conduction parameters and serologic data (such as total serum cholesterol level, LDL cholesterol level, etc.) (13,29–31). In most studies, the diagnosis of DPN is based on peripheral nerve signal and morphological changes, and these changes are compared with the results of MRN or clinical examination (such as neuropathy symptom score [NSS], neuropathy deficit score [NDS], nerve conduction studies [NCS], etc.) of normal people or patients with nDPN. Furthermore, high-resolution MRN is important for understanding the mechanisms of DPN and developing treatments. Peripheral neuropathy in type 1 diabetes is mainly high signal on T2WI, while type 2 diabetes is associated with the increase of T2WI low-signal lesions, and the increase in both lesion types was associated with an increase in disease severity (29). A correlation was found between T2WI-hyperintense lesions and nerve conduction function and HbA_1c, and a correlation was observed between T2WI-hypointense lesions and serum triglyceride and serum HDL. Hence, DPN of type 1 diabetes may be related to poor glycemic control, and that of type 2 diabetes may be related to dyslipidemia (28). It has also been found that patients with painful DPN have more T2-weighted hyperintense nerve lesions than individuals without pain or with no DPN, and their lesions extend over a wider area (29). This suggests that pain sensory symptoms of diabetic polyneuropathy are associated with the proximal fascicular damage.

Currently, PSD (13,26,30,32) and the T2 relaxation time, or T2 value (13,26,30,32,33), are frequently used in DPN imaging. These studies involved patients with type 1 or type 2 diabetes, and the examination site was mainly the sciatic nerve or tibial nerve (TN). By measuring and comparing the PSD or T2 values of peripheral nerves in DPN patients, patients with diabetes but without polyneuropathy (nDPN), and healthy controls, the diagnosis of DPN and verification of the diagnostic efficiency of the sequence was achieved. A study Vaeggemose et al. (26) included 49 patients with type 1 diabetes (11 with severe polyneuropathy [sDPN], 13 with mild polyneuropathy [mDPN], and 25 with nDPN) and 30 healthy control individuals. The results showed that the PSD between nDPN and sDPN was different in the sciatic nerve and TN. Another study by Vaeggemose et al. (32) included 21 patients (11 with DPN and 10 with nDPN) and 10 healthy people. The results showed that PSD could not detect neuropathy. The different results of the two studies may be caused by the different compositions of mDPN and sDPN. There are differences in whether PSD can be used in the diagnosis of DPN. Some conclusions (26,30) have shown significant diagnostic effectiveness of PSD, and it could be a new quantitative imaging biomarker for early DPN (30). Others (13,32) concluded that PSD could not detect DPN. Kronlage et al. (17) found that PSD was negatively correlated with BMI. The increased effect of neuropathy on PSD and the reduced effect of obesity may offset each other. This view may explain the differences described above. Therefore, when assessing DPN with PSD, it is necessary to control the patient's BMI.

T2 relaxation time is calculated after scanning with the multiecho turbo spin echo sequence, and the T2 value is obtained directly by T2 mapping sequence measurements with almost identical significance.

The use of T2 relaxation time in the diagnosis of DPN seems to be controversial. In two studies on type 1 diabetes DPN by Vaeggemose et al. (26,32), the results showed that T2 relaxation time was less effective in the diagnosis of DPN compared with DTI. Wang et al. (34) found that quantitative measurement of neural T2 value could be used to detect and monitor diabetic neuropathy in diabetic rats. After applying T2 mapping sequence to clinical practice (33), researchers discovered that the T2 value of TN could be used as an alternative and noninvasive quantitative parameter to evaluate DPN. This method calculated that the optimal diagnostic threshold was 51.34 ms.

Therefore, quantitative relaxometry has great potential for the diagnosis of DPN, but it still requires a large sample of experimental research, and it is necessary to evaluate the possible differences between the two types of diabetes.

Diffusion-based techniques have become popular as functional markers in MRI neurography, providing greater diagnostic value than conventional methods (35). We will take DTI as an example to explain the value of these techniques in detail. DTI can accurately assess the structural integrity of the affected nerves (13). FA, AD, RD, and MD are commonly used in DTI and have different meanings in the diagnosis of DPN. FA is a reliable marker of the structural neural integrity of the whole peripheral nervous system in patients with type 1 and type 2 diabetes (21,32,36). Its value is altered by the changes in the microenvironment, the density and diameter of the axon, the density and thickness of the myelin sheath, and the number of myelinated fibers. The FA value decreases when there is axonal damage, demyelination, and nerve edema. AD is determined by the direction in which diffusing molecules are least hindered, as indicated in the tensor by its largest eigenvalue. RD is the average eigenvalue of two smaller tensors. There is evidence that RD represents myelin integrity, whereas AD represents axonal integrity and degeneration (37). In these parameters, we often focus on FA value because the correlation between FA value and clinical, epidemiological, and serological data are generally better than that for RD and MD.

Most studies (21,26,36,38) compare DTI data with NCS examination to evaluate the diagnostic efficacy of DTI. Patients with type 1 or type 2 diabetes underwent DTI studies of their sciatic nerves, common TN, or common peroneal nerves (CPN). FA was positively correlated with nerve conduction velocity (26,36,38) or motor nerve conduction velocity (MCV) (21,39), MD and RD were negatively correlated with MCV, and AD showed no correlation with any electrophysiological parameters (21). The correlation between DTI parameters and NCS confirms that DTI can be used as a diagnostic tool for DPN, and it has shown good diagnostic effectiveness. Patients with DPN have lower FA and higher MD and RD than patients without neuropathy and healthy controls (40). Moreover, DTI measurements show that nerve damage is not only concentrated at the proximal level but also affects the entire nervous system in DPN of type 1 or type 2 diabetes (13,26,36). Xia et al. (21) proposed a cutoff for diagnostic DPN in a DTI study on TN and CPN in patients with type 2 diabetes, namely, the FA diagnostic cutoff of 0.48 for TN and 0.44 for CPN, with good diagnostic performance. This is similar to the values of 0.58 for the TN and 0.59 for the CPN proposed by Wu et al. (39). (Their study subjects included patients with type 1 and type 2 diabetes with DPN.)

NDS and NSS scores are commonly used clinically to evaluate the severity of DPN. Both NDS and NSS were negatively correlated with the FA (26,36,38). Vaeggemose et al. (26) performed DTI on patients with type 1 diabetes with severe, mild, and no DPN. DTI parameters were found to be closely related to the severity of neuropathy. The FA value decreased as the severity of the neuropathy increased, and it was more effective than PSD or T2 relaxation time.

There are few articles about the use of MRI to monitor DPN. Brown et al. (40) used MRI to monitor DPN patients with a 10-week supervised exercise program and found that ADC and FA in TN did not change before and after follow-up, suggesting that MRI cannot detect changes in the microstructure of the TN. We may need to extend the follow-up time for DPN or update the scanning technology or scanning sequence to improve the detection sensitivity.

DPN is generally associated with fiber loss in lower-extremity peripheral nerves, including large myelinated fibers and/or small fibers (41,42). Research on DPN assumes that peripheral nerve damage occurs in the C fibers of the distal lower extremities and gets worse with distance (43). There was a limit of 300 × 500 μm in-plane resolution for MRN imaging, which prevented visualization of fiber types (such as A fibers or C fibers) other than fascicular structures. Individual nerve fibers could not be seen in an MRI scanner with a 9.4- or 14.4-T field of view (25). The majority of MRN studies on DPN were based on basic clinical scores and electrophysiological testing (28,30), so it was hard to determine which types of nerve fibers were affected. Groener et al. (25) reported a correlation between T2WI signal changes in the sciatic nerve MRN and various types of nerve fiber injury in patients with type 2 diabetes and in healthy controls. Both NCS and QST were used to evaluate the function of each type of nerve fiber. The results showed that T2WI MRN lesions of sciatic nerve may be pathophysiologically related to the decline of nerve function of medium and large fibers in DPN. However, it is still impossible to evaluate the function of small fibers.

In the study of Wang et al. (33), patients with diabetes without DPN also had higher T2 values than healthy controls. This suggests that quantitative T2 values have an important potential role in subclinical DPN detection.

In a study of patients with type 1 and type 2 diabetes and prediabetes, Jende et al. (36) found that the decrease of FA in the prediabetes group was not related to age, indicating that the peripheral nerve structural damage associated with diabetes occurred in the early stage of the disease. Furthermore, sciatic nerve fiber bundles are reduced in patients with prediabetes and diabetes, implying that DTI can be used for the early diagnosis of DPN. At present, there is no direct evidence that DTI is used for early diagnosis of DPN. We look forward to the emergence of more studies that correlate MRI data with the results of small-fiber examination and evaluate early DPN.

Whether MRI can be used for the diagnosis of DPN in clinical practice is a question that we need to answer. With the popularity of 3.0-T MRI, more and more hospitals can participate in this kind of study. The need for noninvasive and accurate examinations by patients is also a huge driving force for this research. However, there are several problems to be solved.

First, many studies examine different peripheral nerves, making it difficult to directly compare the MRI parameters of these studies. Therefore, the first step for MRN to be used in the clinic is to unify the scan site. The IENFD is notably higher at the thigh level than at the ankle level in healthy subjects. In DPN, IENFD is decreased both distally and proximally (44). Several recent imaging studies show that the distal sciatic nerve is the site of the earliest and most obvious nerve lesions in DPN (28,36). The sciatic nerve is thick, which is easy to examine by MRI examination and the use of data measurements, and the repeatability of the examination is good. Thus, the distal sciatic nerve level is a good choice for the detection of DPN by MRI.

The second problem is the standardization of MRI equipment parameters and inspection process, which enables the inspection results from different countries and regions to be mutually recognized and compared, thus promoting the development of DPN research around the world.

The third question is whether the patients can tolerate the long MRI scan time. The long-range, multiparametric MRI examination of the peripheral nerve takes a lot of time. Some inspections take nearly an hour. This is a challenge for patients with DPN, especially those with moderate to severe DPN. Moreover, it is difficult to guarantee the quality of MRI images. Therefore, it is also very important to set up the sequence and scope of examination of the patients reasonably. Both MRI and ultrasound can evaluate peripheral neuropathy. How can clinicians choose?

MRN has been shown to be a more reliable method for visualizing lesions than ultrasonography in previous studies (45). Ultrasonography, however, may be more feasible than MRN, since it is less expensive and takes less time to perform. It is excellent for obtaining high-resolution images of superficial peripheral nerves, but it is not able to show deep nerves, nerves surrounded by fat, or nerves beneath bones (46,47), and the severity of DPN cannot be distinguished by nerve ultrasound (48). Both examinations have advantages and disadvantages, so clinicians need to make appropriate choices to meet the diagnostic needs. We have designed a selection process, and detailed information can be found in Fig. 2.

More and more evidence has shown that DPN is not limited to peripheral nerves but extends to the central nervous system. Through MRI, we can understand the structural and functional changes of the central nervous system in patients with DPN. The main sequences applied to DPN research are three-dimensional T1WI, functional MRI, DTI, proton MR spectroscopy, MRI perfusion imaging, and so on. We can use a three-dimensional T1WI sequence to measure the changes in the volume of the brain, gray matter, white matter, and different functional areas of the brain in patients with and without DPN to evaluate changes in brain structure. The main manifestation of DPN patients is atrophy in areas related to somatosensory perception (49,50). Functional MRI is based on the blood oxygen level-dependent (BOLD) effect (51). It can be used to identify and analyze functional connections between remote brain areas (52). In some studies (53,54), it has been found that changes in somatosensory network functional connections can distinguish the irritable clinical phenotype from the nonirritable clinical phenotype, and these studies may provide preliminary evidence for individualized treatment of patients with painful DPN. DTI provides a method for quantitatively measuring the integrity of the white matter tracts of the nervous system. Fang et al. (55) demonstrated that DTI can evaluate the integrity of the axons of the central somatosensory pathways in patients with DPN, and the severity of central axonal dysfunction is related to the severity of peripheral neuropathy. Proton magnetic resonance spectroscopy can provide metabolic information from different body tissues. Selvarajah et al. (56) found that the significant decrease in the thalamic NAA-to-choline ratio in DPN indicated the dysfunction of thalamic neurons, which provided evidence for the participation of thalamic neurons in DPN. Therefore, the study on the central nervous system is equally as important for patients with DPN as it is for patients without it. Our review highlights the application of MRI in peripheral neuropathy of the lower limbs, especially high-resolution MRN, PSD, T2 relaxation time or T2 value, and DTI sequences. We also explored the relationship between T2 value and DPN pathology in animal experiments as well as changes in the central nervous system in patients with DPN. We can conclude that MRI can be used for the diagnosis of DPN, and its usefulness is not limited to peripheral nerves. However, the current research is only the tip of the iceberg, and the potential of magnetic resonance is far beyond what has been seen so far. Diffusion magnetic resonance imaging techniques also include diffusion kurtosis imaging, diffusion spectrum imaging, and intravoxel incoherent motion, among others. In many central nervous system diseases, diffusion kurtosis imaging has been shown to provide more information than conventional DTI (57), but whether this is true for examination of peripheral nerves has not been reported. Other techniques are magnetic resonance perfusion imaging and functional imaging, but their use in examination of peripheral nerves has rarely been reported. We expect that in the future, the technical advantages of MRI can be fully exploited to develop technologies that have faster speed and higher diagnostic efficiency for peripheral nerves.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. X.Z. and F.Z. determined the structure of the review, and each wrote separate sections of the manuscript.