Meta-analysis of arterial spin labeling MRI to identify residual cerebral arteriovenous malformations after treatment

Abstract

Background: To use of statistical methods to assess the diagnostic value of arterial spin labeling (ASL) imaging for follow-up of treated arteriovenous malformations.

Methods: We screened references from four databases, namely, the Cochrane Library, PubMed, Web of Science and Embase, that met the requirements. The methodology quality of the included studies was evaluated using the QUADAS-2 (Quality Assessment of Diagnostic Accuracy Studies-2) tool. Data pertaining to diagnostic performance were extracted, and the pooled sensitivity and specificity were calculated using a bivariate mixed-effects model.

Results: We included six studies with a total of 132 patients with arteriovenous malformation (AVM). The merged sensitivity and specificity of ASL for the diagnosis of brain AVMs with incomplete occlusion after treatment were 0.94[0.86-0.98] and 0.99 [0.59-1.00], respectively. According to the SROC curve summary, the AUC was found to be 0.98 [0.96-0.99]. No significant publication bias was observed.

Conclusion: While ASL does not currently match the diagnostic precision of DSA, it is instrumental in post-treatment surveillance of AVM patients. With the development of ASL technology in the future, this technique holds promise as a minimally invasive diagnostic strategy for AVMs with fewer side effects.

Registration number of prospero: CRD42023422087.

Clinical trial number: Not applicable.

Keywords: Arterial spin labeling; Brain arteriovenous malformations; Meta-analysis.

Fig. 1

Systematic review and meta-analysis study selection flowchart. This flowchart illustrates the systematic process of identifying, screening, and selecting studies for inclusion in a meta-analysis. The right-hand side boxes detail the specific reasons for exclusion, while the left-hand side boxes indicate the number of articles remaining after each stage of the review process

Fig. 2

Methodology quality of 7 included studies generated by review manager 5.4. (a) Risk of bias and applicability concerns graph. (b) Summary of risk bias and applicability concerns. Risk assessments for each study are categorized as “Low Risk” (green), “High Risk” (red), or “Unclear” (yellow). The length of the bars represents the degree of risk across various assessment dimensions

Fig. 3

Forest plots of pooled sensitivity and specificity (Include Huang Yuhao2). In the forest plot, each study is represented by a square whose size is proportional to its statistical weight. Each point in the plot represents an individual study’s sensitivity or specificity estimate. The horizontal lines extending from the squares denote the 95% confidence intervals. Red vertical lines represent the pooled sensitivity and specificity of the overall effect estimate. The forest plot demonstrates a high degree of heterogeneity in sensitivity across the studies

Fig. 4

Forest plots of pooled sensitivity and specificity (Exclude Huang Yuhao2). This updated forest plot presents the sensitivity and specificity data for the included studies after excluding outliers that significantly contributed to high heterogeneity. After excluding data that significantly contributed to high heterogeneity, the analysis revealed reduced heterogeneity, with the final pooled sensitivity and specificity estimates being 0.94 and 0.99, respectively

Fig. 5

Forest plots of pooled sensitivity and specificity (Exclude Huang Yuhao1-2). Following the exclusion of the study by Huang Yuhao (limited exclusively to pediatric populations), the updated forest plot demonstrated a significant reduction in heterogeneity metrics. Subsequent meta-analysis yielded pooled sensitivity and specificity estimates of 0.95 and 0.99

Fig. 6

Forest plots of pooled sensitivity and specificity (Exclude Huang Yuhao and Wu chunxue). The reacquired forest plot includes only studies that used a single radiotherapy regimen for intervention, excluding the studies by Huang Yuhao and Wu Chunxue, which involved multiple treatment modalities. The results show a reduction in heterogeneity, with pooled sensitivity and specificity estimates of 0.97 and 0.99, respectively

Fig. 7

SROC Curve of ASL for Monitoring Treated Cerebral AVM. Each plotted point signifies the sensitivity-specificity pair derived from an individual study, with the red points denoting the aggregated estimates of sensitivity and specificity from the collective analysis of all studies. The area under the curve (AUC), presented as the AUC value, serves as a comprehensive measure of the test’s diagnostic efficacy. The summary operating point indicates sensitivity and specificity of 0.94 and 0.99, respectively. The value of AUC is 0.98 [0.96–0.99]

Fig. 8

Deeks’ funnel plot to test publication bias. Each data point represents a study whose position is based on its effect size estimate and standard error. The regression line shows the expected location of studies without bias. With a p-value of 0.22, which is above the 0.05 threshold, the analysis does not indicate significant publication bias