Deep learning versus radiologists for acute aortic dissection on CT: A systematic review and network meta-analysis.

In 30 seconds

This PRISMA-compliant systematic review and network meta-analysis assessed deep-learning (DL) detection of acute aortic dissection on CT, comparisons with radiologists, and CT segmentation accuracy. Across 10 non-contrast CT detection studies, pooled sensitivity was 0.91 (95% CI, 0.87-0.94) and specificity was 0.90 (95% CI, 0.85-0.94). In five head-to-head studies, DL had higher sensitivity and specificity than the radiologist comparators under experimental reading conditions. The authors judged the evidence low-certainty and the estimates experimental upper bounds, supporting investigation of DL as triage assistance rather than replacement.

Key findings

• For non-contrast CT detection, pooled sensitivity was 0.91 (95% CI, 0.87-0.94) and specificity was 0.90 (95% CI, 0.85-0.94).
• In direct comparisons, radiologist-to-DL relative sensitivity was 0.72 (95% CI, 0.57-0.91) and relative specificity was 0.88 (95% CI, 0.83-0.93).
• Across 16 segmentation studies, pooled Dice was 0.863 (95% CI, 0.795-0.931); true- and false-lumen segmentation performed less well than whole-aorta delineation.

Why it matters

Acute aortic dissection requires timely recognition, and CT-based DL may offer useful detection support. The pooled results suggest potential for triage assistance, while the more difficult lumen-level segmentation highlights that performance depends on the task. Because the evidence is low-certainty and derived largely from case-control datasets with substantial heterogeneity and pronounced small-study effects, the estimates should not be assumed to represent routine clinical performance.

What to keep in mind

Certainty was low: datasets were predominantly case-control, median prevalence was 50%, heterogeneity was substantial, and small-study effects were pronounced. Head-to-head comparisons were limited to five studies and experimental reading conditions.