Samer G. Salman

Introduction Artificial intelligence (AI) is increasingly transforming spine care by enabling analysis of imaging, clinical, and textual data to support diagnosis, risk stratification, and treatment planning. Techniques including deep learning–based image analysis, radiomics, and natural language processing (NLP) offer opportunities to standardize interpretation and derive quantitative biomarkers beyond human visual assessment. This narrative review examines recent innovations in spine-focused biomedical informatics and evaluates their clinical relevance and translational challenges. Methods A narrative review was conducted, focusing on AI in spine care across imaging, radiomics, NLP of electronic health records, and clinical decision support. Literature was selected via targeted PubMed searches; studies describing technical development, validation, and clinical implementation were synthesized, addressing explainability, uncertainty, workflow, and regulation. Results Convolutional neural networks enable vertebral and disc segmentation, detection of degenerative and traumatic pathology, and quantitative assessment of paraspinal musculature. Radiomics approaches extract high-dimensional imaging features that improve the prediction of osteoporosis, fracture risk, tumor characterization, and surgical outcomes. NLP techniques, including emerging large language models, unlock clinically relevant information from unstructured text, enhancing outcome prediction and population-level analyses. Multimodal clinical decision support systems integrating imaging, EHR data, and NLP-derived features demonstrate promise for personalized risk assessment and treatment guidance. Challenges remain regarding data heterogeneity, generalizability, privacy, and prospective validation. Conclusion AI-driven biomedical informatics holds substantial promise to enhance diagnostic accuracy, prognostication, and personalized decision-making in spine care. Successful clinical translation will require robust validation, explainable, uncertainty-aware models, regulatory alignment, and integration into clinical workflows to ensure safe, trustworthy, and patient-centered adoption.

Study Design: Retrospective Cohort Study. Objectives: To determine whether cytochrome P450 2D6 (CYP2D6) metabolizer phenotype is associated with new persistent opioid use after cervical, thoracic, or lumbar spine surgery among opioid-naive adults. Methods: Adults with whole-genome sequencing data who underwent cervical, thoracic, or lumbar fusion or decompression between May 2018 and January 1, 2025, were identified in the All of Us Research Program. Participants had 365 days of baseline observation, 180 days of postoperative follow-up, and no opioid fills from 365 to 30 days before surgery. CYP2D6 phenotype was categorized as poor, intermediate, normal, or ultrarapid metabolizer. New persistent opioid use required a perioperative opioid fill and another fill 90 to 180 days postoperatively. Associations were estimated using Firth-penalized logistic regression adjusted for ancestry principal components and clinical covariates. Results: Among 3,820 participants, 460 (12.0%) developed new persistent opioid use. The overall association across CYP2D6 phenotypes was significant (χ ² = 15.30; P = .002). Compared with normal metabolizers, adjusted odds ratios were 1.48 (95% CI, 1.00–2.18) for poor, 1.05 (95% CI, 0.85–1.30) for intermediate, and 1.50 (95% CI, 0.89–2.54) for ultrarapid metabolizers. Poor metabolizers receiving hydrocodone had higher odds of persistent use (OR, 2.15; 95% CI, 1.07–4.32). Conclusions: CYP2D6 phenotype was associated with new persistent opioid use after spine surgery, with higher risk estimates at the metabolic extremes. Extreme-phenotype estimates were reportable only in European-ancestry participants, so risk in other ancestry groups remains unknown. Medication-specific replication in larger, ancestry-diverse cohorts is needed before clinical implementation.

Background: Before unilateral total knee arthroplasty (TKA), the contralateral radiograph is reduced to a Kellgren–Lawrence grade. We asked what the image adds, what the model reads, and how many views are needed. Methods: A multi-view DenseNet121 survival network predicted 5-year contralateral arthroplasty from routine radiographs, against Cox models over 11 clinical variables with and without the dataset-inferred grade, on a once-read test split. Results: In 741 test patients (106 events), the primary contrast put the combined model’s 5-year censoring-weighted AUROC at 0.844 against the clinical model’s 0.680 among the 740 patients that received both scores (difference 0.164, 0.087 to 0.240). Post hoc, a frontal radiograph alone read 0.837, matching the multi-view arms because 56.6% had only a frontal film; withholding lateral and sunrise views cost 0.078 (0.012 to 0.142) in the 315 patients with both. Grad-CAM put 37.8% of attribution in a joint region covering 25.0%; occluding it left 0.800. Conclusions: The preoperative radiograph adds prognostic information beyond the clinical record. Exploratory analyses place almost all of it in one frontal film and beyond an automated grade, not a radiologist’s read. These internal analyses require external validation.

Study Design: Decision-analytic cost-effectiveness analysis using a Markov state-transition model. Introduction: Lumbar spinal stenosis (LSS) is the most common indication for spine surgery in older adults. Minimally invasive surgery (MIS) tubular decompression and open laminectomy produce equivalent one-year outcomes, but their long-term cost-effectiveness for pure decompression is unknown. We modeled 10-year cost-effectiveness from the U.S. payer perspective. Methods: We built a five-state Markov model for a simulated cohort of 10,000 patients aged 65 with pure LSS (no instability, deformity, spondylolisthesis, or prior surgery), using 3-month cycles, a 10-year horizon, and 3% annual discounting. QALYs used SPORT EQ-5D utilities. Perioperative complication rates came from the Nerland 2015 propensity-matched cohort (10.6% vs. 14.6%; risk difference 4.1 percentage points, 95% CI -2.6 to 10.7). Recurrence, postoperative utility, and perioperative mortality were equalized across arms. One-way, threshold, and probabilistic sensitivity analyses (PSA) were performed, including wide-uncertainty sampling of the complication difference across its confidence interval. Results: MIS yielded 5.95 QALYs at $26,283 versus 5.91 QALYs at $30,450 for open laminectomy, saving $4,167 per patient (MIS dominant). MIS was cost-effective in 81% of PSA iterations at $50,000/QALY and in 82% to 88% under wide-uncertainty sampling, and remained cost-effective at $100,000/QALY unless its complication rate exceeded 17.5%. Without any complication advantage, MIS still saved $3,454 with equal QALYs. Conclusion: For pure LSS, MIS tubular decompression was cost-effective and, in the base case, dominant over open laminectomy. The advantage reflects the complication difference and the lower index cost, and should inform rather than determine coverage policy.