In this study, we have investigated the distribution of PDFF in human cadaver vertebrae, with a particular focus on the transition zone between the cancellous and cortical regions of the bone (PCZ). As a flexible anchoring device would be affixed in the PCZ along the outer edge of the vertebra, the area around the spinal cord was considered irrelevant for the given application and was thus not part of the analysis.
The cadavers included in this study had a mean age of 77.8 years and belonged to the patient cohort of older adults, which is the most common cohort for spinal fusion surgery. Patients in this cohort may suffer from back pain due to various clinical conditions including degenerative disk disease and spinal stenosis [2]. Measurements of PDFF in the cancellous bone from this ex-vivo study are in line with the results of previous in vivo research on similar patient cohorts [13, 19, 20].
Similar PDFF distributions and mean PDFFs were observed for all cadavers, except for Cadaver 5. The study results show lower PDFF for Cadaver 5 compared to the rest of the cadavers studied. This subject was found to suffer from malignant neoplasm of the esophagus, which may be the cause of the low PDFF found across all spinal levels. Patients with active malignancy have a higher chance of perioperative complications and are less likely to be considered for spinal fusion surgery [21, 22]. The PDFF measurements for Cadaver 5 can therefore be assumed to be non-representative of vertebral body fat fraction in spinal surgery patients. The mean PDFF difference between cortical bone and PCZ1 found for this cadaver suggests that guidance based on fat fraction may still be possible for patients with active malignancies, but parameters would have to be assessed separately for these patients.
For the other cadavers, the observed PDFF distributions suggest that cortical bone can be distinguished from the remaining ROIs. Fat fraction seems to increase gradually from cortical bone through PCZ1 up to the three innermost ROIs (PCZ2, PCZ3, and cancellous bone). As the PDFF distributions of these three ROIs overlap, no distinction based on PDFF measurements seems possible here.
Statistical analysis confirms these findings: when examining the mean PDFF difference of PCZ3 vs. cancellous bone, no significance is found, as both positive and negative values are observed. Equally, for the mean PDFF difference of PCZ2 vs. PCZ3, the observed values do not consistently have the same sign, hence these zones are not considered significantly different.
When advancing from PCZ2 towards PCZ1, a first significant drop in the mean PDFF can be observed. For the examined samples, the average difference was between −7.59 pp (Fig. 2a) and −4.39 pp (Fig. 2f). Although consistently negative, the absolute values of the observed differences are small for some vertebrae, and it needs to be verified whether they can reliably serve for guidance in spinal fusion surgery.
When further advancing from PCZ1 towards cortical bone, another significant decrease in the mean PDFF is found. For the examined samples, the average difference was between −27.09 pp (Fig. 2a) and −18.96 pp (Fig. 2c). This decrease is in the same order of magnitude as the total mean cortical bone PDFF, and can, therefore, very likely be detected intra-operatively, and thus prevent the surgeon from traversing the cortical bone boundary.
For singular vertebrae, the PDFF curves reveal unusually large deviations from the whole spine mean. These vertebrae also show an altered anatomy on the PDFF MR images. Modic changes that come along with degenerative edema can lead to elevated grayscale values, which are associated with a high PDFF [23]. Another cause for large deviations from the whole spine mean are sclerotic lesions, which can, for instance, manifest as bone islands—intramedullary condensations of cortical bone which appear as areas with low signal intensity.
In Cadaver 3, which reveals a particularly high variation in PDFF (Fig. 1c), several vertebrae exhibit dark spots. Possible explanations are an underlying malignancy with metastases that have destroyed the bone partially, or posterior vertebral scalloping that is a possible result of a variety of pathologies such as degenerative spine conditions, dural ectasia, and intraspinal tumors deforming the vertebra [24]. The influence of such anomalies on lipid content in the vertebrae needs to be researched, although the mean differences acquired in this study do not reveal substantial discrepancies for vertebrae with an altered anatomy.
Furthermore, it has been shown previously that the PDFF changes over the course of a lifetime [25]. This study focused on the most common patient cohort of older adults, and none of the examined cadavers belonged to the other patient cohort of adolescents suffering from spinal deformities [1]. A further study investigating fat fraction distribution in the vertebrae of this patient cohort is encouraged.
It could be argued that the vertebral fat content of cadavers may not represent the in vivo fat content due to postmortem changes. However, a study by Lamoureux et al. [26] showed that bovine and equine percentage of fat in bone marrow does not change within 30–60 days after necropsy, regardless of the storage condition. In an in vivo human study by de Boer et al. [27] fat content was assessed on tissue samples both before and after resection. Comparison of the measurements did not yield any significant differences.
Limitations
The model used for water–fat separation assumes that objects are scanned at body temperature. Although this model is relatively stable to variations in temperature, the calculated PDFF values might be slightly biased, as the cadavers examined in this study were not scanned at body temperature but at room temperature.
Selection of the vertebra contours was done by manually detecting high grayscale values on the MR image. Although the process was kept consistent for the entire dataset, it is prone to bias. Using CT images as ground truth for vertebra contour detection is recommended for future studies. Another possible approach to mitigate the bias is to increase the magnetic field strength from 1.5 T to 3 T for better separation of fat and water, thereby creating a higher contrast between cortical and cancellous bone on the MR images [28]. Image acquisition with an increased in-plane resolution could decrease the pixel size and thus increase the number of data points for each ROI.
The cortical bone ROI was grown automatically based on the detected vertebra contour and the assumption of a uniform cortical thickness of 1 mm. Swamy et al.[29] have shown cortical bone thickness to vary between 1 and 3 mm. However, cortical bone at a distance of more than −1 mm from the vertebra contour is expected to show an equal or lower PDFF compared to the cortical bone ROI as defined in this study, creating an even larger mean difference between cortical bone and PCZ1.
Lastly, investigating PDFF distributions across additional slices and other 3D planes could provide further insights, especially concerning the PDFF distribution in the pedicle area, a crucial region for screw placement.