Introduction: Avoidance of interbody subsidence is essential to achieve adequate indirect foraminal decompression following standalone lateral lumbar interbody fusion (LLIF). Pre-operative risk stratification is limited to bone density evaluation via DEXA or computed tomography (CT) Hounsfield units (HU) region of interest. Utilizing 3D-segmentation on pre-operative CT-imaging we evaluated vertebral bodies and their respective endplates to identify characteristics correlated with endplate subsidence.
Methods: A retrospective analysis of 30 standalone LLIF segments, 17 patients, was conducted at a single institution. DICOM files were segmented using an open-source machine learning algorithm and analyzed on Python 3.9.6. Segmented HU-matrixes were sectioned in axial half and rotated so interbody-adjacent endplates were orthogonal to the x-axis. Averaging HU across the sagittal plane, a saggital –appearing heatmap displays HU density. Further averaging over the coronal plane results in line graph depicting HU change over distance from disc space. Endplate thickness was defined as distance to inflection point between the endplate and trabecular HU. Post-operative implant subsidence was graded according to Marchi criteria. Endplate mean HU, endplate thickness(mm), and minimum axial trabecular HU was compared between subsided interbodies and non-subsided interbodies using the Wilcoxon sign rank test.
Results: Six surgical levels experienced subsidence. There was no significant difference in subsidence rates between patient sex, BMI. Subsided patients were older (71±5.9 vs. 59±7.2). Surgical levels that subsided had lower minimum trabecular HU for upper (179 vs 251, p=0.0119) and lower vertebral bodies (179 vs 260, p=0.003). Subsided levels had lower average HU for upper (340 vs 460, p=0.0023) and lower endplates (327 vs. 447, p=0.005), and the lower endplates were thinner (2.5mm vs 4.1mm, p=0.016). Upper end plate width did not differ (p=0.16).
Conclusion : Machine learning assisted 3D-segmentation can identify unique risk factors for subsidence following standalone LLIF. Thinner caudal endplates and lower vertebral body and endplate HU are risk factors for interbody subsidence.
.jpg)
