Retrospective feasibility; small external set (n=40); quantitative accuracy measured in non-LVO patients, only qualitative rating in the LVO target population; no clinical endpoint; authors state prospective validation needed.
Sources: Primary — https://doi.org/10.1007/s00234-026-04098-2 (Neuroradiology)
- nnU-Net model reconstructing the complete cerebral arterial tree from NCCT in LVO-AIS patients for thrombectomy planning, published online 2026-07-03; industry co-authors from UnionStrong (Beijing) Technology Co., Ltd (academic lead: First Affiliated Hospital, Nanjing Medical University).
- Data: dataset 1 n=280 (training + internal validation + internal test) and dataset 2 n=40 (external validation), both paired NCCT–CTA in patients WITHOUT LVO (Dec 2018–Jul 2025); dataset 3 n=290 LVO-AIS patients from two hospitals, segmentation verified against post-recanalization DSA by two radiologists.
Motivation: CTA cannot opacify vessels distal to an occlusion; the model infers the expected arterial anatomy from NCCT as complementary input for planning (not a contrast/CTA replacement).
- Quantitative performance (on non-LVO sets): DSC 0.80±0.04 (internal validation), 0.79±0.04 (internal test), 0.79±0.04 (external validation, n=40).
- In the LVO target population (n=290): no DSC reported — radiologist-rated "high-quality" in 98.9% (Rater 1) / 98.3% (Rater 2), inter-rater Cohen κ 0.74 (95% CI 0.56–0.89); a subjective quality/agreement measure, not accuracy.
- Evidence quality: retrospective feasibility; no downstream DTN/mRS/clinical endpoint.
Workflow & Architecture
- Position: inference on the first NCCT — could add complementary vascular information before thrombectomy planning (whether it removes any CTA/contrast step is untested).
- Integration: NCCT-in → arterial map out; of potential interest to sites without 24/7 CTA/contrast capability (unvalidated in that setting).
- Friction: clinicians must trust a synthetic vascular map for real decisions — high validation bar.
Competitive / Regulatory / Commercial
- Pushes the "NCCT-only advanced imaging" frontier — the strategic race to extract more from the cheapest, most available scan.
- No clearance/reimbursement event.
- Potential access/equity angle if validated: vascular context where CTA isn't routine.
#MobileStrokeUnit #Neuroradiology #StrokeAI #AcuteStrokeReadyHospitals #PrimaryStrokeCenter #Telestroke #Neurology