Apparent diffusion coefficient maps in the assessment of surgical patients with lumbar spine degeneration

Purpose To assess the utility of apparent diffusion coefficient (ADC) maps for the assessment of patients with advanced degenerative lumbar spine disease and describe characteristic features of ADC maps in various degenerative lumbar spinal conditions. Methods T1-weighted, T2-weighted and diffusion weighted (DWI) MR images of 100 consecutive patients admitted to the spinal surgery service were assessed. ADC maps were generated from DWI images using Osyrix software. The ADC values and characteristic ADC maps were assessed in the regions of interest over the different pathological entities of the lumbar spine. Results The study included 452 lumbar vertebral segments available for analysis of ADCs. Characteristic ADC map features were identified for protrusion, extrusion and sequester types of lumbar disk herniations, spondylolisthesis, reactive Modic endplate changes, Pfirrmann grades of IVD degeneration, and compromised spinal nerves. Compromised nerve roots had significantly higher mean ADC values than adjacent (p < 0.001), contralateral (p < 0.001) or adjacent contralateral (p < 0.001) nerve roots. Compared to the normal bone marrow, Modic I changes showed higher ADC values (p = 0.01) and Modic 2 changes showed lower ADC values (p = 0.02) respectively. ADC values correlated with the Pfirrmann grading, however differed from herniated and non-herniated disks of the matched Pfirrmann 3 and 4 grades. Conclusion Quantitative and qualitative evaluation of ADC mapping may provide additional useful information regarding the fluid dynamics of the degenerated spine and may complement standard MRI imaging protocol for the comprehensive assessment of surgical patients with lumbar spine pathology. ADC maps were advantageous in differentiating reactive bone marrow changes, and more precise assessment of the disk degeneration state. ADC mapping of compressed nerve roots showed promise but requires further investigation on a larger cohort of patients.


Methods
T1-weighted, T2-weighted and diffusion weighted (DWI) MR images of 100 consecutive patients admitted to the spinal surgery service were assessed. ADC maps were generated from DWI images using Osyrix software. The ADC values and characteristic ADC maps were assessed in the regions of interest over the different pathological entities of the lumbar spine.

Results
The study included 452 lumbar vertebral segments available for analysis of ADCs. Characteristic ADC map features were identified for protrusion, extrusion and sequester types of lumbar disk herniations, spondylolisthesis, reactive Modic endplate changes, Pfirrmann grades of IVD degeneration, and compromised spinal nerves. Compromised nerve roots had significantly higher mean ADC values than adjacent (p < 0.001), contralateral (p < 0.001) or adjacent contralateral (p < 0.001) nerve roots. Compared to the normal bone marrow, Modic I changes showed higher ADC values (p = 0.01) and Modic

Introduction
Low back pain is an important socioeconomic and health problem of the modern society and a major cause of disability in adults of working age [1]. The most frequent causes of low back pain include various degenerative pathological conditions of the lumbar spine, for example lumbar disk herniations, spondylolisthesis, degenerative disk disease, many of which have indications for surgery when the conservative treatment fails to provide relief. Magnetic resonance imaging (MRI) is widely used for imaging evaluation of patients with low back pain. T1 and T2-weighted MRI sequences (T1-weighted and T2-weighted) provide anatomical information regarding the soft tissues primarily; including fat and water content and are usually used for the assessment of the lumbar discovertebral complex. Evaluation of T1 and T2 sequences is focused on the structural changes in the intervertebral disk and zygapophyseal joints, reactive vertebral bone marrow changes, location and extent of the disk material displacement, degree of the stenosis, location and extent of nerve root compression. Diffusion weighted imaging (DWI) is a standard MRI sequence for the brain; however, it is not often used for imaging of the spine. A series of DWIs taken at different b values, a measure of the gradient magnetic field strengths, can be used to calculate an apparent diffusion coefficient (ADC) map, which shows the relative speed by which water can diffuse through a tissue. Nutrient diffusion limitation has been elucidated as an important pathophysiological mechanism of disk degeneration, contributing significantly to biomolecular and cellular changes that develop as part of the degenerative pathology [2,3]. Besides its use for intervertebral disk, assessment of the tissue diffusion phenomenon may improve our understanding of other normal and pathological features of degenerating spine and bring new insight to the diagnosis and treatment of degenerative spinal disorders [2,3]. Some specific uses of ADC maps in regards to discovertebral complex include analysis of bone mineral density of lumbar vertebrae [4], diagnosis of a lumbar vertebral chordoma [5], or response of myeloma to treatment [6].
Previous studies focused on the assessment of ADC maps in healthy IVDs [7] or age [8] and degeneration [9] related ADC changes. Such studies were performed on healthy volunteers or patients seen in outpatient settings [7]. The purpose of this study was to retrospectively assess the utility of ADC mapping in pre-operative patients with various known spinal conditions and to determine if they may provide information that is clinically relevant to the preoperative assessment of lumbar degenerative spinal disease. Additionally, we quantitatively assessed ADC maps of the patients with vertebral bone marrow changes and compromised neve roots.

Study population
The study protocol was approved by the "Local Ethics Committee of the Irkutsk Scientific Center of Surgery and Traumatology". All patients provided written voluntary informed consent for their participation in the study. Pre-operative lumbar spine MR imaging was used to investigate 100 consecutive patients who were operated on for various pathologies in the lumbar spine. Patient' characteristics are presented in the Table 1. Study included all consecutive patients scheduled for the surgery on the lumbar spine. We excluded patients with the spinal pathologies at the levels other than lumbar, tumors and vascular pathology.

MRI parameters
Imaging was performed on a 1.5T Siemens Magnetom Essenza scanner (Siemens Healthineers, Erlangen, Germany). Sagittal T1-weighted imaging (WI), T2-WI, and DWIs were collected for Image processing ADC maps were generated from the series of DWI images using the "ADCmap" plugin (http:// web.stanford.edu/~bah/software/ADCmap) in Osyrix (R) software. Sagittal ADC maps, T1-WI and T2-WI were reviewed and visual qualitative patterns on ADC maps that may be valuable for the assessment of degenerative pathology of the lumbar spine were defined. ADC maps were displayed on a full dynamic scale with the Jet color look-up table. Additionally, 65% transparent ADC maps were placed over the corresponding T2-WI for better appreciation of anatomical localization. To quantify diffusion from the lesional and normal spinal structures, the regions of interest (ROI) were drawn over various anatomical locations on sagittal ADC maps and mean and standard deviation (SD) ADC values were calculated. Oval ROIs of 40 mm 2 size were used to assess diffusion in IVD. Mean ADCs from the center of the nucleus pulposus quantified diffusion, and the SD characterized homogeneity of the nucleus. ADC maps were assessed in various types of lumbar disk herniations (protrusions, extrusions, sequestrations [10]), spondylolisthesis, Modic changes [11], Pfirrmann disk degeneration grades [12], central stenosis of the spinal canal, and also from the compressed spinal nerve roots. During the measurement of ADC from IVD, special attention was paid to place the ROI within the limits of the nucleus pulposus on the ADC maps. In a few cases with significantly collapsed disk ROI was flattened and reduced to minimum 30 mm 2 in order to fit the limits of the disk space. ADC measurements on the nerve roots were performed by the raters blinded to the information about the clinically affected nerve root. Image review and measurements of ADC values were performed by 3 neurosurgeons specialized in spinal surgery with 5-15 years of experience and familiar with generation and interpreting of ADC maps. Three neurosurgeons collectively evaluated all images on the same database and agreed on the ROI locations for the ADC measurements.

Statistical analysis
Statistical analysis was performed in Statistica (Dell Inc., Tilsa, OK, USA) software. Data presented as means and standard deviations. Nonparametric tests were used to calculate coefficients of significance p: Wilcoxon matched pairs test (W), Kruskal-Wallis ANOVA (K-W), Mann-Whitney U test (U), and Spearman rank R correlation. Differences were considered significant when p < .05.

Results
ADC maps were generated successfully for all patients. The CSF signal had a mean ADC of 2752±333 x10 -6 mm 2 /s. The ADC maps were calculated for the lower lumbar levels from L3 to S1 in all patients, L2-3 in 93/100 (93%) patients, L1-2 in 59/100 (59%) patients, resulting in 452 vertebral segments available for diffusion analysis. Eight segments were excluded from the analysis due to the image artifacts (bending of the DWIs image signal that changed the ADC values significantly). Metal spinal implants introduced imaging artifacts and so corresponding segments were excluded from the analysis (Fig 1).

Types of lumbar disk herniations
Disk herniations (n = 114) were classified as protrusion (n = 67), extrusion (n = 34) or sequester (n = 13) type based on the T2-WIs. ADC/T2-WI overlay images provided additional visual information about the composition of disk tissue. Sagittal ADC maps allowed for a visual differentiation of the protrusions composed of bulging annulus fibrosus (Fig 2), from the protrusions with annular clefts and protruded nucleus (Fig 3). ADC maps increased contrast and allowed for better differentiation of annular clefts and protruding nucleus when such distinction was not clear on a T2-WI.   Extruded herniations were clearly visible on both T2-WI and ADC maps. Interestingly, ADC/T2-WI composites were helpful for characterization of the extruded tissue structure and for defining the extent of nucleus pulposus extrusion under the posterior ligaments (Fig 4).
Recently extruded sequesters retained mean ADC value of the nucleus pulposus ( Fig 5); however, in some older sequesters the mean ADC value of the nucleus pulposus increased. For instance, in one representative example, sequester ADC value was higher than the ADC value from the adjacent non-herniated L3-4 nucleus pulposus (1665 ± 381 vs. 1449 ± 273 x1 -6 mm 2 /s, p < 0.05). Among the all herniated disks (n = 114, all Pfirrmann grades included in analysis), there were no differences in the mean ADC values measured on the center of the disk when compared between the various LDH types (p K-W = 0.13), various LDH locations (p K-W = 0.74) and various LDH levels (p K-W = 0.77).
Modic type 1 changes were evident on the ADC maps (Fig 8). Mean ADC values from the Modic type 1 changes associated with degenerative spondylolisthesis were higher than from the vertebral bone marrow. Modic type 1 changes were associated with significantly higher ADC values compared to a similar ROI from the adjacent healthy vertebral bone (498 ± 139 vs. 314 ± 86 x 10 −6 mm 2 /s, respectively, p U = 0.01) and when compared to Modic type 2 changes (223 ± 110 x10 -6 mm 2 /s, p U < .001) ( Table 2).
Modic type 2 changes representing bone marrow replacement by a fat were visually similar and had significantly lower ADC values compared to the intact vertebral bone on ADC maps (p U = 0.002) (Fig 9). All segments with type 2 changes showed low ADC values from both the disk area and the bone marrow. We did not observe significant changes that could be interpreted as Modic type 3 representative of subchondral bone sclerosis in the assessed patients.

Disk degeneration
Quantitative assessment showed that the mean ADC values have a significant correlation with the Pfirrmann degeneration grades (Fig 10, S1 Table). Visual grading of highly degenerated disks on ADC maps was not straightforward due to the inhomogeneity of the disk. Disks with Pfirrmann grade 4 and 5 were significantly less homogeneous than grade 2 and 3 (Fig 11). Among the grade 4 disks the mean ADC values were significantly higher in herniated compared to non-herniated disks (p U < 0.01), and opposite for the grade 3 disks: herniated disks had significantly lower mean ADC values than non-herniated (p U = 0.04) (Fig 12). Herniated disks of grades 3 and 4 showed significantly more ADC heterogeneity than non-herniated disks of the same grades (p U < 0.01 and p U = 0.01 respectively) (Fig 13).

Correlation analysis
Mean nucleus ADC values inversely correlated with spinal level, Pfirrmann grade, age, and presence of disk herniation. Nucleus ADC heterogeneity showed opposite correlations of slightly lower power with the same variables (Table 3).

Discussion
In this study we retrospectively investigated ADC maps both qualitatively and quantitatively in a selective cohort of 100 patients who underwent surgery for the treatment of degenerative lumbar spine pathology. To our knowledge, this is a first study that focuses on assessment of ADC maps of the degenerative pathology of the lumbar spine. On a large patient sample this study described characteristics of ADC images for different types of disk herniations, nerve root compression, reactive endplate changes and disk degeneration grades.  Qualitative features of ADC maps ADC maps overlaid onto T2-WIs provided additional contrast and information about the diffusivity of different types and locations of disk herniations. There were no overall significant differences in mean ADC values from the nucleus in different types and locations of herniations; however, this study did not investigate local changes in ADC values of different parts of the disk. Additionally, we found association between increase in diffusion heterogeneity and disk tissue degeneration. Future studies focused on the assessment of local distributions of ADCs across the various parts of the disks and adjacent structures are required to validate our qualitative visual observations in different pathological conditions of the IVD. Nerve roots. DWIs and ADC maps of spinal nerve roots, ganglions, and nerves with generation of tractography [13,14] were shown to add value for the preoperative assessment of spinal nerve root lesions [15,16]. However, ADC values from the dorsal root ganglion in symptomatic patients were found to be inconsistent, but correlate with recovery from pain and numbness [17]. Furthermore, not all symptomatic patients have increased ADC values in the regions of compromised nerve roots and ganglions. Interestingly, our study showed significant differences in ADC values from the symptomatic nerve roots compared to the nerve roots  of the contralateral and adjacent segments, although visually ADC maps were not strikingly different. In patients with clinical signs of monosegmental radiculopathy, ADC values from the compressed nerve root were significantly higher than any neighboring nerve root (ipsilateral adjacent, contralateral on the same level, or contralateral on the adjacent level) in 17/19 (89%) patients. Only in 2 patients the adjacent ipsilateral nerve root had higher ADC values than the affected nerve root. If compare nerve roots on symptomatic level only, ipsilateral nerve root ADCs (1300±221 x 10 −6 mm 2 /s) were on average 25% higher than contralateral (1038±118 x 10 −6 mm 2 /s) in all 19/19 patients, p W < 0.001. Furthermore, ADC values from  the ipsilateral adjacent nerve roots were significantly higher compared to the contralateral nerve roots on the same level (p W < 0.048), and nearly significantly higher compared to contralateral nerve roots on the adjacent level (p W = 0.09). Such findings are in accordance with the frequent clinically observed ipsilateral symptoms in the LDH patients [18,19]. Only rarely, does contralateral radiculopathy occur with LDH [20]. Additionally, we have reviewed intraoperative videos of selected cases with monosegmental radiculopathy and found that ADC values may be increased in compressed nerve roots without the typically assumed visual appearance of inflammation during surgery (Fig 14). Our findings confirm previously reported preliminary findings with a comparable number of patients [21,22]. However, we believe that further larger study is warranted to determine the specificity and sensitivity of ADC maps for the assessment of radicular symptoms and nerve root compression. Sagittal ADC maps were used in this study for the nerve root assessment, while previous studies used axial ADC maps [15,16]. Axial ADC maps may be easier to identify nerve roots, but weather the quantitative ADC measurements from the axial and sagittal scans are different remains unknown. At this time, our data form 26 patients suggest that increased ADC values alone would have a high positive predictive value, but low sensitivity for the diagnosis of the o. male M. with left sided herniated disk at L5-S1 level. Intraoperative picture after removal of the herniation shows slightly edematous S1 nerve root (arrow). Left S1 nerve root ADC value is 1030 ± 192 x1 -6 mm 2 /s. ADC, apparent diffusion coefficient; LDH, lumbar disk herniation.
https://doi.org/10.1371/journal.pone.0183697.g014 symptomatic spinal nerve root compression. Additionally, low visual contrast between the compressed and healthy nerves and surrounding tissues is still a limiting factor for the use of ADC maps. Modic changes. This study showed that Modic type 1 changes are associated with increased ADC values, which corresponds to an increased diffusion in this region compared to the vertebral endplates without Modic changes or to the Modic type 2 changes. This finding reflects the inflammatory, edematous [23], or infectious [24] nature of the pathological substrate in Modic 1 changes. The probable "claw sign" [25] was found in all segments with the Modic type 1 changes, favoring degeneration over infection. In 5/51 (10%) segments with Modic 2 changes we have found high mean ADC values similar to what was seen in Modic type 1 changes. These emphasized the complexity of the underlying mechanisms of reactive vertebral body bone marrow changes, which do not always represent fat degeneration [26]. Modic changes were predominantly found in grade 4 and 5 degenerated disks with lower ADC values.
Pfirrmann grades. This study has also revealed potential limitations of the Pfirrmann grading system for the assessment of degeneration in herniated disks. The mean nucleus ADC values and the ADC heterogeneity differed significantly between the herniated and non-herniated disks in the matched Pfirrmann grades. Also, transition in ADC values from grade 3 to grade 4 is quite abrupt. We suggest that Pfirrmann grades 3 and 4 may cover an overly broad range of states of disk degeneration. Some herniated disks were assigned Pfirrmann grade 4 because of a dark inhomogeneous signal on the T2-WI. However, ADC values were significantly higher than expected for this Pfirrmann grade 4 because when the nucleus pulposus dislocates, the center of the disk contains less hydrated nuclear tissue and more annular and transition zone tissue. Indeed, the original grading scale developed by Boos and Pfirrmann [12] was described without specific attention to herniations, lysthesis, endplate changes, or discitis. Whether the freshly herniated disks cause temporary increase in local diffusion and increased ADC values remains unconfirmed. Injury to the avascular annulus fibrosis would likely not result in edema to the same extent as other more vascularized tissues. Another explanation for increased ADC values in the degenerated disk is a presence of cracks filed with an exudate [7]. Proper grading of disk degeneration is important for early diagnostics, comparison across clinical studies, and estimation of regenerative potential. Therefore, more adequate grading systems should be considered [27][28][29]. ADC maps may detect changes not visible on T2-WI [8] and therefore may be used as an adjunct for more precise evaluation of the disk degeneration.
Advantages and disadvantages of ADC maps. Overall, the benefits of ADC maps for the assessment of patients with surgical pathology of the lumbar spine include several aspects. First, the imaging of the spinal nerves may provide additional information on their precise anatomical location, possible entrapment, and potential functional prognosis. Second, ADC maps allow for more precise estimation of IVD degeneration. Third, ADC/T2-WI composites provide increased image contrast and additional functional information for the assessment of the normal and pathological anatomy of the lumbar spine. Finally, vertebral bone marrow and endplate infectious, degenerative and fatty changes could be differentiated more easily with ADC maps.
The drawbacks of diffusion imaging include the lack of the anatomical image, artifacts on the edges of the scan area, and possible patient movement during scanning. Another limitation is the additional time required for DWI sequence acquisition and ADC map processing. Automated calculation and overlay of ADC maps on the corresponding anatomical images, such as T2-WI, will make this modality more useful in the clinical settings.
Limitations. There are several limitations in this study. The study population represents a selected subgroup of spine surgery patients with lumbar disk herniations, spondylolisthesis and symptomatic lumbar disk degeneration. Therefore, generalization of the findings to a larger population of patients without such pathology should be done with caution. The number of patients with lysthesis was low, but we assume that the qualitative ADC map will be similar for other cases, although this should be confirmed in a larger cohort. The imaging was performed on a 1.5 T MRI with defined TE, TR, and b values. Direct comparison of the ADC values obtained with other imaging parameters may differ and should be done with caution. Calculating interobserver variability was not within the scope of this study. We deemed for this study it was better to have a discussion amongst the expert surgeons with regard to ROI placement, grading, and interpretation of herniation.

Conclusions
This study revealed benefits and limitations of qualitative and quantitative ADC map evaluation of pre-operative lumbar spine patients using a contemporary 1.5 T MRI. ADC maps overlaid on T2-WIs provide functional information about tissue diffusion, which may be useful for assessment of the morphology of various degenerative conditions in the lumbar spine. ADC maps were advantageous for differentiating reactive vertebral bone marrow changes and for more precise assessment of the disk degeneration state. ADC mapping showed promise for detection and assessment of compressed nerve roots but this aspect requires further investigation on a larger cohort of patients.
Supporting information S1