AI for Deep Vein Thrombosis Detection: Scoping Review of 11 Studies Finds Promise but Gaps

Scoping review of 11 studies (2021–2025) finds AI detects deep vein thrombosis with ultrasound sensitivity 68–100% and specificity 70–100%; MRI-based models reached 95–97% across metrics. Authors conclude AI suits a supplementary role — multicenter validation still lacking.

• Scoping review across 7 databases; 11 eligible studies evaluated AI (RetinaNet, deep neural networks, and others) applied to CT, MRI, and ultrasound for DVT diagnosis — ultrasound models were most studied.
• MRI-based models achieved sensitivity, specificity, and accuracy of 95–97%; the single CT-based model reported sensitivity of 83%; multi-dataset models reached precision ≥96% (CIs and p-values not reported in source).
• Key limitation: only 11 small, predominantly single-center studies with no prospective multicenter external validation; QUADAS-2 quality assessment flagged methodological heterogeneity across studies.

RadPigeon summaries are original and for information only. They are not clinical advice.