The diagnosis of ASD with MRI: a systematic review and meta-analysis

Summary & key facts

This review combined 134 studies (159 experiments) that tried to tell apart people with autism spectrum disorder (ASD) from healthy controls using resting-state MRI. Across studies the pooled accuracy was moderate: about 76% sensitivity, 75.7% specificity, and an AUC of 0.823. However, the authors say there is a lot of variation between studies, overlap in samples, and uncertainty about how well results would work in real clinics, so confidence in the exact numbers is limited. They conclude MRI-based methods show promise but face important obstacles before they can be used reliably in practice.

Key facts:

• 134 peer-reviewed studies were included, representing 159 eligible experiments.
• The estimated number of unique participants across studies was 4,982: 2,439 people with ASD and 2,543 healthy controls.
• Pooled sensitivity was 76.0% (95% CI 74.1–77.8).
• Pooled specificity was 75.7% (95% CI 74.0–77.4).
• The pooled area under the curve (AUC) was 0.823, a summary measure of diagnostic performance.
• Only resting-state MRI studies were included; task-based fMRI studies were excluded to reduce heterogeneity.
• The search covered publications from January 1, 2018 to December 31, 2022 and used PubMed and Web of Science (last searched Feb 24, 2023).
• The authors reported major limitations: substantial heterogeneity between studies, overlap in samples, and uncertainty about whether the reported performance would generalize to new clinical settings.

Topics

Autism Spectrum Disorder Research Child Development and Digital Technology Genetics and Neurodevelopmental Disorders

Categories

Cognitive Neuroscience Life Sciences Neuroscience

Tags

Autism Autism spectrum disorder Bivariate analysis Computer science Confidence interval Diagnostic accuracy Internal medicine Law Machine learning Magnetic resonance imaging Medicine MEDLINE Meta-analysis Pathology Political science Psychiatry Psychology Publication bias Radiology Study heterogeneity