UCLA's DSA-NRP predicts stroke no-reflow instantly from angiograms
New ML model predicts microvascular hypoperfusion right after thrombectomy, skipping 24-hour MRI delay.
After successful large-vessel recanalization via endovascular thrombectomy (EVT) for acute ischemic stroke, up to some patients experience 'no-reflow' — persistent microvascular hypoperfusion that undermines tissue recovery. Current clinical practice relies on perfusion MRI within 24 hours post-procedure, delaying critical intervention. A new paper from UCLA researchers introduces DSA-NRP, the first machine learning framework to predict no-reflow immediately after EVT using intra-procedural digital subtraction angiography (DSA) sequences.
The retrospective study analyzed 2011–2024 AIS patients with favorable mTICI scores (2b-3) who had pre- and post-procedure MRI. From DSA (AP and lateral views), the team extracted statistical and temporal perfusion features from the target downstream territory to train ML classifiers. DSA-NRP significantly outperformed a clinical-features baseline (AUC: 0.7703 ± 0.12 vs 0.5728 ± 0.12; accuracy: 0.8125 ± 0.10 vs 0.6331 ± 0.09), demonstrating that real-time DSA perfusion dynamics encode critical microvascular integrity insights. This approach eliminates the need for 24-hour delayed MRI, enabling proactive management of high-risk patients and potentially improving stroke outcomes.
- DSA-NRP uses intra-procedural DSA sequences (AP and lateral views) to predict no-reflow after EVT.
- Model achieves AUC 0.7703 and accuracy 0.8125, significantly beating clinical-only baselines (AUC 0.5728).
- Eliminates the 24-hour post-procedure MRI delay, enabling immediate risk assessment and intervention.
Why It Matters
Immediate no-reflow prediction could reduce stroke disability by enabling timely microvascular treatment without MRI delay.