External validation of a top-ranked model from the RSNA pulmonary embolism detection challenge: assessment of generalizability.
An AI tool detected major blood clots in the lungs with 94% accuracy, though it struggles with smaller clots requiring human judgment.
The RSNA 2020 challenge 2nd-place DL algorithm achieved AUROC 0.94 for any PE detection on 1,038 independent CTPAs (sensitivity 0.80, specificity 0.97, accuracy 88.7%), with excellent central PE performance (AUROC 0.99) but poor subsegmental PE detection (AUROC 0.48). This record was retained from the prior triage attempt for PubMed pipeline handoff.
What the study was
- Study design
- External validation study (retrospective CTPA cohort, n=1,038)
- Category
- ai_ml_diagnostics
- Maturity
- Validated
Why it surfaced
Real-world external validation of a high-profile competition DL model for PE; directly relevant for clinical radiology deployment decisions. Quantifies performance gaps (subsegmental PE weakness) that matter for patient safety and regulatory approval.
A plain-language summary of published research — not medical advice. Talk to a clinician about your own care.