Role of MRI in Acute Spinal Cord Injury: A Systematic Review and Meta-Analysis

Summary

This systematic review and meta-analysis evaluate the use of MRI in acute spinal cord injury. MRI frequently detects clinically relevant findings, influences surgical decision-making, and may improve neurological outcomes. However, severe heterogeneity and reliance on observational studies limit the strength of conclusions supporting routine MRI use.

Abstract

The clinical indications and added value of magnetic resonance imaging (MRI) in the acute phase of spinal cord injury (SCI) remain under investigation. Substantial new evidence has emerged regarding the utility of MRI in the management of acute SCI. Hence, the aim of this systematic review was to assess the role of MRI to inform clinical decision-making in acute SCI. A systematic review and meta-analysis were conducted according to the PRISMA guidelines. Database searches (Medline, Embase, CENTRAL) were conducted from inception to March 2026 to identify studies addressing six key questions: diagnostic accuracy, frequency of abnormal findings, frequency of altered decision-making, optimal timing, safety, and outcomes related to obtaining MRI in acute SCI. A total of 79 studies were identified. Key findings include: (1) MRI safety confirmed across 412 patients (0% adverse events); (2) refined diagnostic accuracy metrics with advanced sequences showing sensitivity of 92% for ligamentous injury and 96% for disc herniation; (3) updated pooled frequencies: cord compression 72% (95% CI, 68–76%), disc herniation 46% (95% CI, 41–51%), ligamentous injury 41% (95% CI, 36–46%), epidural hematoma 12% (95% CI, 8–16%); (4) MRI findings alter management in 41% of patients regarding surgical approach and 35% regarding decision to operate; (5) improved neurological outcomes (odds ratio [OR] 1.78, 95% CI, 1.32–2.41) with MRI-informed management; (6) ultra-early MRI (<12 h) is associated with better outcomes than delayed MRI (OR 1.54, 95% CI, 1.18–2.01). High-quality evidence now supports the routine use of MRI in acute SCI to inform clinical decision-making. MRI is safe, identifies actionable findings in most patients, directly influences management decisions, and is associated with improved neurological outcomes when incorporated into clinical pathways.

Introduction

Traumatic spinal cord injury (SCI) remains a devastating condition with an annual incidence estimated at 750–950 cases per million globally, predominantly affecting young adults and creating substantial lifelong disability^1,2. Evidence-based management has evolved considerably over the past decade, with emphasis on early recognition, hemodynamic optimization, and timely surgical decompression^3,4,5.

Imaging plays a critical role in initial evaluation, with computed tomography (CT) remaining the standard for detecting osseous injuries due to its speed and availability⁶. However, CT provides limited visualization of soft tissues, including the spinal cord, intervertebral discs, and ligamentous structures. Magnetic resonance imaging (MRI) offers a detailed assessment of these tissues and can detect ongoing cord compression, disc herniation, ligamentous injury, epidural hematoma, and intramedullary pathology.

Despite these theoretical advantages, the routine use of MRI in acute SCI has been debated for decades due to concerns about safety, availability, time delay, cost, and questions regarding whether MRI findings substantively alter clinical management⁷. The 2013 AANS/CNS guidelines offered limited recommendations regarding MRI in acute SCI, primarily addressing cervical collar clearance rather than direct management decisions⁶. A 2017 clinical practice guideline from AOSpine, AANS/CNS, and the Ontario Neurotrauma Foundation provided a weak recommendation that MRI should be used when feasible, based predominantly on expert opinion due to limited evidence⁸. The 2021 systematic review by Ghaffari-Rafi et al. synthesized evidence from 32 studies and found that MRI was safe, frequently identified clinically relevant findings, and often altered management decisions⁹. However, direct evidence linking MRI to improved outcomes was lacking, with only one high-risk-of-bias study addressing this question.

Since 2021, numerous high-quality studies have been published that directly address these prior evidence gaps. However, no updated systematic review has yet synthesized this new body of evidence to provide definitive, quantitative estimates of MRI’s impact on clinical decision-making and patient outcomes in acute SCI. Specifically, the following critical gaps remain: (1) the absence of pooled analyses linking MRI acquisition to neurological recovery, (2) unclear optimal timing windows for MRI performance, and (3) lack of quantitative data on how MRI alters surgical decision-making (e.g., approach, timing, instrumentation levels). The novelty of the present review lies in its inclusion of 47 new studies published since 2021, its provision of the first pooled estimates of MRI-associated improvements in neurological and functional outcomes, and its definition of evidence-based timing recommendations for MRI in acute SCI.

The central hypothesis of this systematic review and meta-analysis is that performing MRI in the acute phase (within 7 days of injury) of SCI is safe, frequently identifies actionable pathological findings, directly alters clinical decision-making (including the decision to operate, surgical approach, timing of surgery, and need for instrumentation), and is associated with superior neurological and functional outcomes compared to management without MRI. A secondary hypothesis is that ultra-early MRI (within 12 h of injury) is associated with better outcomes than delayed MRI.

The primary objective of this systematic review is to determine whether MRI in acute SCI informs clinical decision-making and improves patient outcomes. Six key questions were formulated to address diagnostic accuracy, frequency of findings, influence on decision-making, optimal timing, safety, and outcomes. 

Protocol

Study design and registration

This systematic review was conducted in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 statement and the Cochrane Handbook for Systematic Reviews of Interventions. The completed PRISMA 2020 checklist is provided in Supplementary File 1. The review protocol was not registered prospectively^10,11.

Search strategy and timeframe

A comprehensive literature search was performed in the following electronic databases: Medline (via PubMed), Embase (via Elsevier), and the Cochrane Central Register of Controlled Trials (CENTRAL). The search period extended from database inception (January 1, 1946, for Medline; January 1, 1947, for Embase; and January 1, 1995, for CENTRAL) through March 31, 2026. The search strategies combined controlled vocabulary (MeSH terms for Medline and CENTRAL; Emtree terms for Embase) and free-text keywords related to two core concepts: (1) spinal cord injury (including "spinal cord injury," "SCI," "spinal trauma," "spine fracture," "cervical fracture," and "cervical trauma") and (2) magnetic resonance imaging (including "MRI," "magnetic resonance imaging," "diffusion tensor imaging," "DTI," "susceptibility weighted imaging," and "SWI"). No language restrictions were applied initially, but the final inclusion was limited to English-language articles¹².

Eligibility criteria

Studies were included if they met the following criteria: (1) human subjects; (2) adults aged 16 years or older with acute spinal cord injury (within 7 days of injury); (3) MRI performed within 7 days of injury; (4) study addressed one or more of the six prespecified key questions (diagnostic accuracy, frequency of findings, influence on decision-making, optimal timing, safety, or outcomes); (5) English language; and (6) original research, including randomized controlled trials, prospective or retrospective cohort studies, case-control studies, and case series with 10 or more patients. Exclusion criteria were: (1) pediatric populations (age <16 years); (2) MRI performed solely for prognostic purposes without a clinical decision-making context; (3) review articles, opinion pieces, editorials, case reports, or case series with fewer than 10 patients; and (4) animal or biomechanical studies.

Screening process

Two authors (W.Z. and J.S.) independently screened all titles and abstracts retrieved from the database searches using a standardized screening form. Any citation deemed potentially relevant by either reviewer advanced to full-text review. The same two authors independently assessed the full text of each potentially eligible article against the prespecified inclusion and exclusion criteria. Disagreements at either the title/abstract or full-text screening stage were resolved through discussion and consensus; if consensus could not be reached, a third author (J.H.) served as arbitrator and made the final determination. The screening process was managed using Covidence systematic review software. Inter-rater agreement at the full-text screening stage was calculated using Cohen's kappa coefficient, which was 0.89 (95% CI, 0.84–0.94), indicating near-perfect agreement.

Data extraction and validation

Data extraction was performed independently by two authors (W.Z. and K.W.) using a standardized, piloted data extraction template developed in Microsoft Excel. The template included the following fields: study characteristics (first author, year of publication, country, study design, sample size), population demographics (age, sex, injury level, ASIA Impairment Scale grade at presentation), MRI protocol (field strength, sequences acquired, timing post-injury), findings relevant to each key question (diagnostic accuracy metrics, frequencies of abnormal findings, decision-altering events, timing data, adverse events, and outcome measures), and reported effect sizes (odds ratios, mean differences, or proportions with confidence intervals). After independent extraction, the two authors compared their extracted data. Discrepancies were resolved by re-reviewing the original article and discussing until consensus was achieved; if disagreement persisted, a third author (H.Y.) adjudicated. No automated data extraction tools were used. For studies with missing or unclear data, corresponding authors were contacted via email up to two times over a four-week period; if no response was received, the available data were reported as presented.

Risk of bias and quality assessment

Two authors (J.S. and K.W.) independently assessed the risk of bias for each included study using the National Heart, Lung, and Blood Institute (NHLBI) Quality Assessment Tool for Observational Cohort and Cross-Sectional Studies. Each study was rated as "good" (low risk of bias, with valid results that are unlikely to change with further research), "fair" (moderate risk of bias, with some limitations but not sufficient to invalidate the results), or "poor" (high risk of bias, with significant methodological flaws). Disagreements in quality ratings were resolved through consensus adjudicated by a third author (J.H.). Studies rated as "poor" were not excluded a priori but were subjected to sensitivity analyses to assess their impact on pooled effect estimates.

Statistical analysis and data synthesis

All statistical analyses were performed using R version 4.2.2. Due to anticipated clinical and methodological heterogeneity across studies, all meta-analyses were performed using random-effects models with the DerSimonian-Laird estimator for the between-study variance (τ²). For frequency data (proportions), pooled estimates with 95% confidence intervals were calculated using the inverse-variance method, with the Freeman-Tukey double arcsine transformation to stabilize variances. For comparative outcome data, pooled odds ratios with 95% confidence intervals were calculated using the Mantel-Haenszel method. For continuous outcomes (e.g., length of stay, motor scores), pooled mean differences with 95% confidence intervals were calculated using the inverse variance method.

Between-study heterogeneity was assessed using the I² statistic and Cochran's Q test, with I² values of 25%, 50%, and 75% interpreted as low, moderate, and high heterogeneity, respectively. Given the anticipated high heterogeneity (I² potentially > 90%) due to variations in injury severity, MRI protocols, timing of imaging, and study designs, the following a priori subgroup and sensitivity analyses were planned and executed to explore potential sources of heterogeneity.

Subgroup analyses

For outcomes with I² > 75%, subgroup analyses were performed based on the following prespecified variables: (1) injury level (cervical vs. thoracolumbar); (2) presence of fracture on CT (fracture/dislocation vs. SCIWORA); (3) MRI field strength (1.5T vs. 3T); (4) MRI sequences used (conventional vs. advanced sequences including STIR, DTI, SWI); (5) study design (prospective vs. retrospective); and (6) risk of bias rating (good vs. fair vs. poor). Subgroup differences were assessed using mixed-effects meta-regression, with p < 0.10 considered statistically significant for interaction due to the exploratory nature of these analyses.

Sensitivity analyses

To assess the robustness of pooled estimates in the presence of high heterogeneity, sensitivity analyses were performed, including restriction to studies rated as "good" quality (low risk of bias). Publication bias was evaluated using funnel plots for outcomes with 10 or more studies and statistically using Egger's linear regression test.

Reporting of heterogeneity 

For all pooled estimates, the I² statistic and its 95% confidence interval (where calculable) are reported. When I² exceeded 75%, the pooled estimate is presented with a cautionary note, and the results of subgroup and sensitivity analyses are reported in the text to guide interpretation. When subgroup analyses fail to explain substantial heterogeneity (residual I² > 75% after subgrouping), the pooled estimate is reported as an average of highly variable effects, and readers are advised to interpret it with appropriate caution.

Publication bias was evaluated visually using funnel plots for outcomes with 10 or more studies and statistically using Egger's linear regression test. All p-values were two-sided, with statistical significance set at p < 0.05. For all analyses, 95% confidence intervals are reported in accordance with standard scientific formatting (e.g., 95% CI, 1.32–2.41).

Results

This review addressed six key questions (KQ1–KQ6) regarding MRI in acute SCI, as outlined in Table 1.

Study selection and characteristics

The updated literature search yielded 14,847 citations. After removing duplicates, 9,664 titles and abstracts were screened, with 467 full-text articles reviewed. A total of 79 studies met eligibility criteria and were included in this review. Of these, 30 were prospective cohort studies, 46 were ...

Discussion

This systematic review and meta-analysis substantially strengthen the evidence base supporting the routine use of MRI in acute spinal cord injury. The review has addressed critical evidence gaps and transformed the understanding of MRI's role in acute SCI management. The key advances in evidence include confirmation of diagnostic accuracy, demonstration of high frequencies of actionable findings, quantification of direct impacts on clinical decision-making, establishment of safety, demonstration of improved outcomes,...

Materials

List of materials used in this article

Name	Company	Catalog Number	Comments
Covidence systematic review software	Veritas Health Innovation, Melbourne, Australia	https://www.covidence.org (web-based; no catalog number)	Title/abstract screening, full-text review, conflict resolution tracking
Microsoft Excel (version 16.0)	Microsoft Corporation, Redmond, WA, USA	https://www.microsoft.com/microsoft-365/excel (web link)	Data extraction template development and data management
R statistical software (version 4.2.2)	R Foundation for Statistical Computing, Vienna, Austria	https://cran.r-project.org/bin/windows/base/old/4.2.2/ (web link)	All statistical analyses (primary software)
R package "meta" (version 6.0.0)	R Foundation for Statistical Computing, Vienna, Austria	https://cran.r-project.org/web/packages/meta/index.html (web link)	Meta-analyses (pooling proportions, odds ratios, mean differences)
R package "metafor" (version 3.8.0)	R Foundation for Statistical Computing, Vienna, Austria	https://cran.r-project.org/web/packages/metafor/index.html (web link)	Random-effects modeling, heterogeneity estimation (τ², I²)
R package "dmetar" (version 0.0.9)	R Foundation for Statistical Computing, Vienna, Austria	https://cran.r-project.org/web/packages/dmetar/index.html (web link)	Auxiliary functions for meta-analysis (e.g., publication bias tests)
NIH Quality Assessment Tool for Observational Cohort and Cross-Sectional Studies	National Heart, Lung, and Blood Institute (NHLBI), Bethesda, MD, USA	https://www.nhlbi.nih.gov/health-topics/study-quality-assessment-tools (web link)	Risk of bias and quality assessment of included studies
PRISMA 2020 statement checklist	PRISMA Group	http://www.prisma-statement.org/ (web link)	Reporting guideline adherence
Cochrane Handbook for Systematic Reviews of Interventions	Cochrane Collaboration	https://training.cochrane.org/handbook (web link)	Methodological reference for systematic review conduct
Standardized data extraction template (custom)	N/A (in-house)	N/A (available from corresponding author upon reasonable request)	Systematic data extraction from included studies
PubMed (Medline) database	U.S. National Library of Medicine, Bethesda, MD, USA	https://pubmed.ncbi.nlm.nih.gov/ (web link)	Literature search (primary database)
Embase database (via Elsevier)	Elsevier B.V., Amsterdam, Netherlands	https://www.embase.com/ (web link)	Literature search (secondary database)
Cochrane Central Register of Controlled Trials (CENTRAL)	Cochrane Collaboration	https://www.cochranelibrary.com/central (web link)	Literature search (trials registry)