Introduction: Osteoporotic spine patients are becoming more prevalent. Dual-energy X-ray absorptiometry (DEXA) is the standard for bone mineral density (BMD) evaluation; although, lumbar and DEXA scores are inaccurate or extrapolative in degenerative spinal pathology. Current clinical applications of computed tomography (CT) HU unit analyses are limited to region of interest analysis. We utilize a machine learning (ML) algorithm for 3D-image segmentation for evaluation of osteoporosis through comparison with DEXA correlates.
Methods: A retrospective analysis of 290 patients with CT and DEXA scans within 1-year of each other was performed at a single tertiary institution (2014-2024). CT scans were segmented using ML and cumulative HUs for lumbar vertebra were generated and compared to their DEXA T-scores. Utilizing DEXA scores, the Youden index identified ideal cutoff values representing osteoporosis and osteopenia. Segmentation and analyses were performed with Python 3.9.6 and SAS 9.4.
Results: 290 patients (85% female) with a mean age of 72±13 years and BMI of 28±7.2 kg/m2 were analyzed. With DEXA categorization from lumbar T-scores, 119 were osteopenic (41%) and 37 were osteoporotic (13%). Compared to normal patients, average individual lumbar level HUs were lower in osteoporotic patients (L1-L5 all p< 0.005) and in osteopenia patients (L1-L5 all p< 0.008). Total lumbar HU mean was significantly lower in osteopenia and osteoporotic patients (normal, 248±46; osteopenia, 227±62; osteoporosis 197.9±60; p< 0.0001). Lumbar DEXA score strongly correlated with lumbar average HU (r 2=0.25, p< 0.0001). Average lumbar HU was predictive of DEXA defined osteopenia (ROC 0.74, p< 0.0001) and osteoporosis (ROC=0.77, p< 0.0001). Optimal average lumbar HU threshold value to identify DEXA-defined osteoporosis was 196 HU (OR 6.4, p< 0.0001) and osteopenia was 220 HU (OR 6.3, p< 0.0001).
Conclusion : An automated ML-powered segmentation and HU analysis of CT imaging can match DEXA for the diagnosis of osteoporosis in spine patients. This approach could spare spine patients costly DEXA scans enabling opportunistic screening and dual use of CT images.