Three-dimensional distribution of CT attenuation in the lumbar spine pedicle wall

This study investigated in vivo the three-dimensional distribution of CT attenuation in the lumbar spine pedicle wall measured in Hounsfield Unit (HU). Seventy-five volunteers underwent clinical lumbar spine CT scans. Data was analyzed with custom-written software to determine the regional variation in pedicle wall attenuation values. A cylindrical coordinate system oriented along the pedicle’s long axis was used to calculate the pedicular wall attenuation distribution three-dimensionally and the highest attenuation value was identified. The pedicular cross-section was divided into four quadrants: lateral, medial, cranial, and caudal. The mean HU value for each quadrant was calculated for all lumbar spine levels (L1–5). The pedicle wall attenuation was analyzed by gender, age, spinal levels and anatomical quadrant. The mean HU values of the pedicle wall at L1 and L5 were significantly lower than the values between L2–4 in both genders and in both age groups. Furthermore, the medial quadrant showed higher HU values than the lateral quadrant at all levels and the caudal quadrant showed higher HU values at L1–3 and lower HU values at L4–5 than the cranial quadrant. These findings may explain why there is a higher incidence of pedicle screw breach in the pedicle lateral wall.


Subjects. A total of 84 volunteers participated in this IRB-approved study (Rush University Medical Center
Institutional Review Board, No. 00042801) including all subjects signing an approved informed consent form. All research was carried out in compliance with relevant guidelines and regulations. From this initial recruitment group, nine of them were excluded for the following reasons: one volunteer had spondylolisthesis, seven volunteers had transitional lumbosacral vertebrae and the last one had a combination of spondylolisthesis and transitional lumbosacral vertebrae. In consequence, a total of 75 subjects were used for the analyses.
Imaging. Each subject underwent CT imaging of the lumbar spine (L1-S1, Volume Zoom, Siemens, Malvern, PA, tube voltage: 120 kV, tube current: 100 mA, field of view: approximately 200 mm, image matrix: 512 × 512, slice increment: 1.0 mm, slice thickness: 1.0 mm, no spacing) in a supine position. Axial slice raw image data were exported in the DICOM format.

Measurement of attenuation in the vertebral body. Attenuation in the vertebral body was measured
using the software ImageJ (US National Institutes of Health, Bethesda, MD) to evaluate osteoporotic changes of the vertebral bodies. For each of vertebral bodies the largest possible elliptical ROI was drawn for the vertebral body HU measurement 28 . These ROIs were acquired on all the axial CT slices of the L1-L5 lumbar levels. Osseous abnormalities and voids such as vascular channels were excluded from the ROIs.
Determination of local density distribution in 3D space adjacent to the pedicle axis measured in Hounsfield Units (HU). To obtain the 3D distribution of HU values in the space adjacent to the pedicle axis, a cylindrical discrete coordinate system centered along the pedicle long axis was defined as follows: Approximate center points in both the posterior (point a) and anterior (point b) ends of each pedicle were manually determined in the DICOM image(s) of the pedicle level using commercially available software (Mimics R21, Materialise Corp., Leuven, Belgium) (Fig. 1A,B). Point c was set on line ab and moved from a to b with in increments of 1/30 of the length of line ab . This enabled the definition of vector − → cd in a cranial direction perpendicular to vector − → ab . Point d revolved 360° in 6° increments on a plane perpendicular to ac and a probe point e was determined. The magnitude of vector − → cd was increased from 0.5 mm to 13.0 mm in 0.25 mm increments. The HU value at each point e was calculated by a trilinear interpolation algorithm using HU values at 8 adjacent pixels in 2 adjacent axial CT slices (Fig. 1C) 29 . A total of 90,000 HU datapoints were obtained within the cylindri-

Results
In total, n = 750 lumbar pedicles (L1-5) from 75 volunteers (39 males and 36 females; mean age, 39. 3 Table 2). There were no significant differences between levels in total, gender and age groups. Two subjects (one was 45 years old male; one was 23 years old female) had the vertebral attenuation values at L1 less than 110 HU, which was proposed as a cut-off yielding high specificity for osteoporosis by Pickhardt et al 31 .
The peak HU values in the pedicle wall were symmetric between both right and left sides. Considering spinal level, there were no significant differences by gender with the exception of L1 (p < 0.05) (Fig. 4, Table 3). The mean HU values in the 20 s/30 s group were higher than those in the 40 s/50 s (Fig. 5, Table 3). The mean HU values at L1 and L5 were significantly lower than the values between L2-4 in both genders and age groups (p < 0.03) (Figs. 4, 5, Table 3). From L1 to L3 the HU values followed this pattern: Lateral < Medial < Cranial < Caudal. The HU values in the lateral region were lower than those in the medial region at all spinal levels (p < 0.007) (Fig. 6, Table 3). The HU values at the caudal region were higher than those in the cranial region at L1, L2 and L3 (p < 0.0009, p < 0.0001 and p < 0.0001, respectively), however the HU values in the cranial region were higher than those in the caudal region at L4 and L5 (p < 0.003, p < 0.0001, respectively) ( Fig. 6, Table 3). Pairing the pedicle wall HU and the vertebral HU by spinal level showed no correlation except at L4 in females and at L2 in males (Table 4). In the age group of 40's + 50's, these correlations were found at L2 and L4 (Table 4).  www.nature.com/scientificreports/

Discussion
Although previous studies reported on the pedicle cortical bone thickness 7,19-23 and the attenuation values of the whole pedicle or trabecular bone inside the pedicle 32-34 , the present study is the first to evaluate in vivo the regional attenuation distribution of the lumbar spine pedicle wall in all lumbar levels using clinical CT data. This report showed that there was an increase in HU values of the pedicle wall from L1 to L2, followed by a consistent plateau of data between L2 and L4 and a decrease at L5 in both gender groups and both age groups. Our regional distribution of attenuation showed that the lateral region had lower HU values than the medial region at all spinal levels. We also showed that HU values in the caudal region were higher than in the cranial region from L1 to L3, but lower at L4 and L5. The increase seen in the average HU values from L1 to L2-4 may be attributed to a corresponding gradual rise in loading at the lower levels. However, the average HU values decreased between L4 and L5, which is consistent with results of previous research on the whole lumbar pedicle HU values, including the pedicle cortex and interior trabecular bone 32 . This finding may be explained by remarkable changes in pedicle dimensions between L4 and L5. Sugisaki et al. reported a significant increase in longitudinal axis of the pedicle isthmus from L4 to L5 by 25% and 27% in females and males, respectively 30 . Assuming pedicle cross-sectional geometry has an Attenuation split by spinal level and age. *p < 0.04 between age groups, **p < 0.009 between age groups, ***p < 0.002 between age groups, ****p < 0.0006 between age groups, (a) p < 0.0001 compared with L2, L3 and L4 and p < 0.03 compared with L5 in the same age group, (b) p < 0.0001 compared with L2, L3 and L4 in the same age group, (c) p < 0.003 compared with L2 and p < 0.0001 compared with L3 and L4 in the same age group, (d) p < 0.02 compared with L2 and p < 0.002 compared with L3 and L4 in the same age group. compared with cranial (p < 0.003) and caudal (p < 0.0001) in L1, (c) p < 0.0009 compared with caudal in L1, (d) compared with medial (p < 0.006), cranial (p < 0.0001) and caudal (p < 0.0001) in L2, (e) p < 0.0001 compared with cranial and caudal in L2, (f) p < 0.0001 compared with caudal in L2, (g) compared with medial (p < 0.0002), cranial (p < 0.0001) and caudal (p < 0.0001) in L3, (h) p < 0.0001 compared cranial and caudal in L3, (i) p < 0.0001 compared caudal in L3, (j) compared with medial (p < 0.0001), cranial (p < 0.0001) and caudal (p < 0.001) in L4, (k) compared with caudal (p < 0.05) in L4, (l) p < 0.003 compared with caudal in L4, (m) compared with medial (p < 0.007, cranial (p < 0.0001) and caudal (p < 0.004) in L5, (n) compared with caudal (p < 0.0001) in L5, (o) p < 0.0001 compared with caudal in L5. www.nature.com/scientificreports/ elliptical shape, these increases in the longitudinal axis would increase area moment of inertia to resist a bending moment in females and males by 95% and 105%, respectively. Therefore, the reduced HU values in L5 may be of limited influence on pedicular structural properties, leading to consider that a reduction of pedicular HU values in L5 may be a result of functional adaptation of bone tissue under reduced stress caused by the increased area moment of inertia.

Scientific Reports
The lateral region showed lower HU values than the medial region at all spinal levels. It has been reported that lateral wall breaches are the most common pedicle-screw related failures in thoracolumbar indications 12,[20][21][22]35 Previous studies looking at regional variation of the cortical thickness in the lumbar pedicle found that the cortical thickness was thinnest in the lateral region 7,12,21-23 , which has been thought to be a reason for the higher incidence of the pedicle penetration in the lateral pedicle wall. Crawford et al. investigated the trajectory of selfcentering lumbar pedicle screws inserted in trajectories starting 0°, 10°, 20° or 30° from the optimal trajectory, either medially or laterally misdirected. The authors found that lateral misalignment as small as 10° was likely to lead to cortical wall violation (3 of 7 violations) whereas medial misalignment usually resulted in safe screw insertion (1 of 21 violations for 10°, 20°, or 30° medial misalignment) 12 . This study indicates that the lateral wall has less strength to resist pedicle screw penetration as compared with the medial wall. Although multiple factors would be involved in the higher incidence of pedicle screw breach in the lateral side, lower attenuation in the lateral pedicle wall shown in the present study should be considered as an important factor favoring pedicle screw breach due to the strong correlation shown by previous efforts between HU values and cortical bone material properties 36 .
Significant differences in the HU values between the cranial and caudal regions of the pedicle were found in the present study. While the HU values in the caudal region were higher at L1-3, the opposite case was seen at L4-5. This is clinically important in that the inferior breach of the pedicle screw at L4 and L5 is relatively common and can result in neurological complications. The surgeon should be mindful of the cortical walls with lower HU values, which are lateral in all lumbar levels and inferior at L4 and L5. Trabecular tracts running obliquely from the superior process downward to the inferior endplate through adjacent to the cranial cortex of the pedicle and from the inferior process upward to the superior endplate through adjacent to the caudal cortex of the pedicle were reported as early as 1925 by Gallois and Japoit 37 . The architecture of the trabecular bone tract trajectories connecting the articular processes and the lumbar vertebral body 38 and radiographically dense trabecular tracts extending from the pedicles 15 were also reported. Given that these trabecular tracts connecting the articular processes and the vertebral body bear load transmission between the anterior and posterior columns, the pedicle cortex in the cranial and caudal regions may also play a part in load transmission and resist bending moments borne by the pedicle. A full explanation of why are there higher HU values in the caudal region at L1-3 and in the cranial region at L4-5 remains elusive. We speculate that load transmission patterns between the anterior and posterior column through the pedicle, especially bending in the sagittal plane, may be different between the upper and lower lumbar spines, where spinal curvature and posture may play a role. Future biomechanical studies on the stress/strain distribution in the pedicle and their correlation with the HU values distribution will be required to explain the regional differences in pedicle wall HU values.
In the present study, we provided a new cylindrical HU dataset for each pedicle. HU values of new points in the cylindrical dataset were calculated by trilinear interpolation of the eight adjacent HUs in the original CT voxels (Fig. 1C). In the discrete cylindrical coordinate system, we used a smaller voxel size in the discrete cylindrical coordinate system than the original CT voxel size to minimize reduction of the peak HU values associated with trilinear interpolation. It should be noted that the sub-voxel position data of each point in the cylindrical coordinate system were not used for any geometrical analyses of the pedicle in the present study except for quadrant zoning of the pedicle.
Our study is not without some limitations. First, since this study cohort included subjects from 20 to 50 s, who are relatively active and still young, only two subjects met the criteria of osteoporosis defined by the vertebral HU values at L1 31 . Future studies including an osteoporotic population will be required. Second, we analyzed regional variations in pedicle wall attenuation by quadrant regions along the pedicle axis even though the entire 3D HU values distribution data were available. Analyses using other zoning systems and/or pattern analyses www.nature.com/scientificreports/ of the HU values distribution considering cortical bone trajectory would be of interest. Finally, the voxel size of the original CT data used in the present study, approximately 0.4 × 0.4 × 1.0 mm, may not be of high-enough resolution when compared with the pedicle cortical thickness if the intent was to evaluate absolute peak HU values at individual points due to partial volume effects. In summary, this paper presented a novel 3D pedicle model based on peak HU values located radially away from the pedicle axis and investigated regional variation in lumbar pedicle wall attenuation. The results of present study demonstrated that the mean attenuation value at L5 was significantly lower than the values in L2-4 regardless of gender and age and the HU values in the lateral region was lower than in the medical region at all spinal levels. Since the 3D HU values distribution in the pedicle wall can be measured directly from patient-specific DICOM datasets without creating 3D models, the presented method could be used as a preoperative planning for transpedicular fixation using clinical CT.

Data availability
The datasets generated during and/or analyzed during the current study are available from the corresponding author on reasonable request.