Answers are grounded in this paper and cite their sources.
Background/Objectives: Artificial intelligence (AI) is increasingly investigated in scoliosis surgery, but its clinical readiness varies across applications. This structured narrative review examines AI applications across preoperative planning, intraoperative support, and postoperative assessment, focusing on methodological limitations, validation, and clinical utility. Methods: English-language publications were identified iteratively through PubMed and reference-list screening during approximately one year of manuscript preparation. The adopted publication window was January 1, 2022, to August 27, 2026. A.F. and R.K. selected and reviewed the literature. Twenty-two distinct primary studies were organized by clinical task and synthesized narratively. Qualitative appraisal was informed by PROBAST + AI, QUADAS-2, ROBINS-I, CLAIM, and relevant reporting guidelines; formal tool-based risk-of-bias ratings and quantitative pooling were not undertaken. PRISMA-ScR was not applied because this work was designed as a structured narrative review. Results: Preoperative applications included fusion-level selection, prediction of postoperative alignment, and risk stratification. Specialized systems demonstrated promising performance, including external validation, whereas general-purpose multimodal language models showed inconsistent agreement with expert surgical decisions. Applications informing intraoperative care primarily involved pedicle screw trajectory planning and patient-specific rod geometry rather than real-time adaptive guidance. Postoperative applications encompassed automated alignment measurements, instrumentation classification, implant monitoring, and complication prediction. A multicentre radiographic system achieved a mean absolute error of 2.7° for major-curve measurement, while selected growing-rod and vertebral body tethering studies reported measurement errors below 1 mm and 1°, respectively. However, evidence was frequently limited by small retrospective cohorts, insufficient external validation, incomplete calibration assessment, and limited evaluation of clinical impact. Mixed adult spinal deformity cohorts provided indirect evidence and were distinguished from scoliosis-specific populations. Conclusions: AI shows promise for automating defined imaging tasks and supporting individualized scoliosis surgical planning and follow-up. Technical accuracy alone does not establish clinical benefit. Prospective multicentre validation, transparent reporting, and evaluation of effects on decisions, complications, reoperations, and patient-reported outcomes are needed before routine implementation.
No takes yet. Share an insight, caveat, or question.
Kosińska et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: