405 AI Image Analysis of Lumbar MRI to Automate Extraction of Vertebral Bone Quality Score

INTRODUCTION: Bone health is a crucial predictor of spine surgical outcomes. While dual-energy X-ray absorptiometry (DEXA) is the gold standard for measuring bone mineral density (BMD), it is often unavailable preoperatively and can overestimate BMD in patients with spinal degenerative disease and obesity. The MRI-based vertebral bone quality (VBQ) score correlates strongly with DEXA T-scores and predicts osteoporotic fractures and postoperative outcomes, and is an alternative that exposes the patient to less radiation than Hounsfield Unit measurement from CT. However, VBQ calculation is time-consuming, requires expertise, and is subject to variability. To date, no studies have assessed automated VBQ extraction. METHODS: We conducted a multisite retrospective analysis using T1-weighted MRI scans from the SPIDER challenge dataset, involving 257 patients. The dataset was divided into training (n=179), validation (n=39), and test (n=39) sets. On 10 test images the VBQ score was manually calculated. A 2D U-Net convolutional neural network was trained to segment L1-L4 vertebrae and the spinal canal. Average signal intensity of vertebrae (SIL1-L4) was determined by finding the largest oval that fit inside each vertebrae and reducing the radius to exclude cortical bone. SICSF was calculated by using the median intensity in a 1 cm diameter circle in the spinal canal directly posterior to the L3 vertebra. RESULTS: The AI algorithm achieved a mean Dice score of 0.92 (±0.05) for vertebrae segmentation and 0.92 (±0.03) for spinal canal segmentation on the test set. The mean difference of VBQ score with automated vs manual calculation was +0.13 (±0.05). CONCLUSIONS: Our AI-based method significantly improves the accuracy, efficiency, and standardization of VBQ score extraction from MRI’s. If refined, incorporation into electronic health records to create evidence-based risk calculators could identify osteoporotic patients at high risk for poor surgical outcomes.