Routine MRI in Pediatric Scoliosis: A Narrative Review of Risk-Stratified and Abbreviated Protocols.

Magnetic resonance imaging (MRI) is the reference standard for detecting neural axis abnormalities (NAA) in children with presumed idiopathic scoliosis, but whether MRI should be used routinely or selectively remains debated. This SANRA-guided narrative review used structured searches of MEDLINE/PubMed, last run on July 19, 2026, to examine NAA prevalence, the performance of clinical and radiographic predictors, risk-stratified imaging, abbreviated MRI, and the consequences of imaging. Large series and pooled analyses report NAA in approximately 8-10% of presumed idiopathic cases, with higher prevalence in early-onset, congenital, and neurologically abnormal presentations. No single predictor, including left thoracic curvature or abnormal abdominal reflexes, reliably excludes intraspinal pathology. We propose a two-axis strategy that reframes the question from whether to image to when, for whom, and in what form. Axis 1 is PACE (Presentation age, Anatomic etiology, Curve characteristics, Examination), a literature-derived, rule-based, unvalidated framework in which the highest-risk finding determines a high-, intermediate-, or low-risk tier; some thresholds are expert assumptions. Axis 2 is protocol intensity: early full whole-spine MRI, abbreviated first-line MRI, or surveillance with predefined escalation. Evidence for abbreviated MRI is limited to one small retrospective single-center early-onset scoliosis cohort (n = 50), in which limited sagittal sequences identified all 10 NAA detected on full MRI with an estimated 68% shorter acquisition; diagnostic accuracy, anesthesia savings, and cost effects remain unproven. PACE is not a clinical prediction rule or guideline. The conceptual contribution of this review is to separate two decisions that are usually conflated, who should be imaged (risk stratification) and in what form (protocol intensity); both axes, and artificial-intelligence triage, require prospective multicenter and external validation.