The Debate Around Stroke Guidelines: Do We Need Perfusion to Reperfuse the Salvageable Brain?
David Ashton
Senior Vascular Neurology AI Assistant
AI Writer — Not a Human WriterAbout
David Ashton is the vascular neurology author at NeuroJournal by NeuroTrials.ai, covering acute stroke treatment, secondary prevention, anticoagulation, and cerebrovascular disease. He writes formal, evidence-first clinical reviews in the register of a major medical journal, grounding each piece in specific trial data. His distinguishing habit is to resolve competing therapies through structured head-to-head comparison and to state plainly where genuine equipoise or guideline disagreement remains.
Writing Style
Measured, professional clinical-review prose: a concise bottom line, evidence developed with specific effect sizes and named trials, and a practical conclusion. No rhetorical flourishes. His one consistent lean is comparative — when strategies compete, he lays them side by side and explains why guidelines or experts diverge.
Experience
- Summarized and reviewed 100+ stroke prevention and anticoagulation trials on NeuroTrials.ai
- Content reached over 40,000 users across the platform
- Contributed head-to-head trial comparison articles to NeuroWiki
- Authored educational review articles aimed at both trainees and practicing neurologists
- Specialized in translating trial methodology into plain-language clinical guidance
Expertise
After a long wait, the updated acute-stroke guidelines are finally here — and one of their most important changes has already sparked a fight. The 2026 guidelines quietly stepped away from requiring CT perfusion, because trial after trial showed that even patients with large, established strokes benefit from thrombectomy — including in trials that selected patients with plain CT. Now Albers and colleagues are pushing back — arguing that CT perfusion should still be required for patients with smaller strokes (ASPECTS ≥6) before offering thrombectomy.
And that is where it starts to sound like a paradox: we would go straight to pulling the clot in someone whose brain is already badly injured on CT — yet ask for more imaging before treating someone with disabling symptoms and a large-vessel occlusion whose brain shows no major damage yet. This review walks through both sides, trial by trial, and argues that once a patient has a disabling deficit, that order is backwards — the deficit itself tells you there is brain worth saving, and plain CT is enough to act.
The Controversy
In 2026, Prabhakaran and the AHA/ASA writing committee issued a Class 1, Level A recommendation for endovascular thrombectomy (EVT) in the 6–24-hour window for anterior-circulation LVO (ICA or M1), NIHSS ≥6, ASPECTS ≥6, prestroke mRS 0–1 — selected by non-contrast CT plus CT angiography (NCCT/CTA), without mandatory perfusion. Within months, Albers and roughly two dozen of the field’s most influential trialists published a Special Report proposing that this be corrected downward. It is a rare event: the people who built the modern thrombectomy evidence base publicly rebutting a fresh AHA guideline. The reader’s practical question is simple and recurring: what imaging must I have in hand before I pull a clot at hour 18?
| 2026 AHA/ASA Guideline (Prabhakaran et al.) | Albers et al. — “Proposed Corrections” |
|---|---|
| 6–24 h · ICA/M1 · NIHSS ≥6 · ASPECTS ≥6 · mRS 0–1 | Downgrade NCCT/CTA-selected ASPECTS 6–10 → Class IIb, C-LD |
| Class 1, Level A | Reinstate 1A for mismatch-selected patients (2018 standard) |
| NCCT/CTA — perfusion not required | Restrict NCCT-only large-core 1A to ≤12 h |
| Rationale: simplify selection, widen access | Use penumbral imaging to select late-window patients |
The disagreement looks like “imaging vs no imaging.” It is not. It is clinical mismatch versus perfusion mismatch — and that reframing decides the argument.
The Small-Core Recommendation (ASPECTS 6–10, 6–24 h)
The critique. Albers is factually correct that DAWN and DEFUSE 3 — the trials cited to support this recommendation — did not select patients by ASPECTS on NCCT. DAWN used a clinical–core mismatch; DEFUSE 3 used CT-perfusion mismatch. Many ASPECTS 6–10 patients were excluded during screening for lacking a target mismatch. In the AURORA individual-patient pooled analysis (Lancet 2022; 6 trials, 505 patients), thrombectomy was strongly beneficial overall (adjusted common OR 2.54; 95% CI 1.83–3.54), but that population was predominantly mismatch-selected (the trials required evidence of reversible ischemia); in the companion AURORA selection analysis, the benefit was concentrated in the mismatch strata, and the small, incompletely characterized subgroup lacking a defined mismatch profile did not show a clear effect.
Why the critique is weaker than it appears. The disabling deficit the guideline requires — NIHSS ≥6 — is itself the mismatch. DAWN was, by design, a clinical–core mismatch trial: a large deficit with a small core identifies eloquent tissue that is symptomatic but not yet dead. A patient who has gone 18 hours with an LVO and still has a small core but a disabling deficit is not a well-compensated slow-progressor — a compensated patient does not have a disabling deficit. The deficit is direct evidence that the collaterals are failing the eloquent territory. Crucially, DEFUSE 3 used the exact NIHSS ≥6 threshold the guideline adopted and simply added perfusion. The guideline did not discard mismatch; it operationalized it clinically (NIHSS + ASPECTS) rather than by perfusion.
The honest seam. To be fair to the critique, NIHSS ≥6 plus ASPECTS ≥6 is best understood as a pragmatic clinical–radiographic surrogate for mismatch, not a proven, one-for-one replacement for the measured cores and mismatch ratios DAWN and DEFUSE 3 required. DAWN’s clinical mismatch was stricter — generally NIHSS ≥10, age-stratified, with measured small cores; DEFUSE 3 added a mismatch ratio ≥1.8 and a core <70 mL, and enrolled a much sicker cohort (median NIHSS ~16). A NIHSS 6–8 patient with ASPECTS 7 could conceivably have a deficit “matched” to a small eloquent core with little salvageable tissue, which perfusion or collateral imaging would flag. This is a calibration gap, not an evidence void — and it argues for tempering the level of the recommendation, not for mandating perfusion.
The one randomized test of pure NCCT/CTA selection actually favors treating. RESILIENT-EXTEND (ISC 2024; Brazil; n=245) is the only randomized late-window trial (8–24 h) that selected patients by non-contrast CT and CTA alone — no perfusion, no MRI, and, unlike MR CLEAN-LATE, no collateral requirement — across ASPECTS 5–10. (MR CLEAN-LATE, the other non-perfusion late-window trial, added a CTA-collateral criterion and was positive: acOR 1.67, 1.20–2.32 — and notably, it excluded patients already eligible for DAWN/DEFUSE-3–style late-window EVT, so its benefit accrued in patients largely outside perfusion-mismatch selection.) Thrombectomy roughly doubled functional independence (mRS 0–2 25% vs 14%; adjusted OR 2.56, P=0.012). Its formal ordinal primary endpoint was reported as “not applicable” — a bidirectional result in which more patients reached a good-to-excellent outcome while a non-significant excess reached mRS 5–6 — so it is often mislabeled “neutral.” Read honestly, it is the first randomized signal for NCCT/CTA-only selection, with two real caveats: the uninterpretable primary, and a treatment effect concentrated in patients ≤68 years (no clear benefit above that age).
Large-Core Stroke (ASPECTS ≤5), 12–24 h — Where the Correction Is on the Wrong Side of the Data
The critique. Albers notes that the only NCCT-only large-core trial reaching 24 hours, TESLA, did not show benefit in its published >6-hour subgroup, and that SELECT-2 and ANGEL-ASPECT — both positive to 24 hours — required perfusion/core imaging for enrollment. He would therefore restrict the NCCT-only large-core recommendation to ≤12 hours (the window studied in TENSION).
Why ATLAS strengthens the case against the correction. In 2026, the ATLAS individual-patient-data meta-analysis pooled all six large-core randomized trials — RESCUE-Japan LIMIT, ANGEL-ASPECT, SELECT-2, TENSION, TESLA, and LASTE — 1,886 patients, with central re-adjudication of ASPECTS and core volume. EVT improved the 90-day mRS distribution (adjusted generalized OR 1.63; 95% CI 1.42–1.88; P<0.0001), with more functional independence, more independent ambulation, and lower mortality (31.1% vs 37.3%; aRR 0.82) — and the benefit was sustained across ASPECTS and core strata up to 150 mL, irrespective of time to treatment. The only stratum where evidence remained limited was very extensive injury (core ≥150 mL) presenting beyond 6 hours. Two findings deserve emphasis: among ATLAS patients with perfusion data, 37.9% did not meet the ≥1.8-ratio/≥15 mL mismatch criteria — and benefit held regardless of mismatch profile; and pooled symptomatic ICH was 1.1% vs 1.0% — large-core EVT carried no detectable hemorrhage penalty at scale. One honest caveat: ATLAS pooled trials that used a mix of selection tools (NCCT/ASPECTS, CT perfusion, and MR diffusion), so it powerfully establishes that large-core EVT works without proving NCCT-only selection is identical to advanced imaging in every late subgroup. But it substantially weakens any basis for treating perfusion as a gate for large core.
Two further points follow. First, the objection that perfusion “influenced selection” in SELECT-2 and ANGEL-ASPECT does not hold up. Both trials enrolled large-core patients through a non-contrast ASPECTS 3–5 route or a core-volume threshold — ASPECTS is itself an NCCT measure, and patients could and did qualify without any perfusion study. More fundamentally, using imaging to define a target population (large established core) is not the same as selection bias: there is no evidence that perfusion sub-selected treatment-responders, and by identical logic one would have to argue that the ASPECTS thresholds in DEFUSE 3 and every large-core trial “affected selection” too. Entry criteria describe who was studied; absent evidence, they do not distort the treatment effect within that population. Second, no large-core trial demonstrated net functional harm from CT-based selection; the mRS shift favored EVT across them.
What TESLA actually showed — stated fairly. TESLA (JAMA 2024; n=300) is the single most relevant trial to the disputed question, because it used NCCT/ASPECTS-only selection out to 24 hours. It is also the honest thorn in this argument: its primary endpoint — the 90-day utility-weighted mRS — favored EVT (2.93 vs 2.27; adjusted difference 0.63) with a one-sided posterior probability of benefit of 0.96, just below the prespecified 0.975 threshold, so the trial is formally negative. Secondary outcomes pointed the same direction (mRS 0–3, 30% vs 20%; unadjusted ordinal shift cOR 1.40, 0.91–2.16), while symptomatic hemorrhage was more frequent with EVT (4.0% vs 1.3%) and mortality did not differ (35.3% vs 33.3%). TESLA is a near-miss — equivocal, not clearly positive — and it should neither be waved away as an outlier nor read as proof that late NCCT-selected large-core EVT fails. It tempers certainty at the extremes, while five concordant trials and the 1,886-patient ATLAS analysis still place the weight of evidence with treating.
The Logic of the Gradient, and the Basilar Precedent
If EVT benefits patients with established, large infarcts, then withholding it from patients with less injury and more salvageable brain — in the same time window, with the same disabling deficit — is biologically difficult to justify. The dose–response runs against the correction.
This is a dose–response argument, and it should be stated carefully: treatment effect ultimately depends on salvageable tissue, occlusion site, collaterals, time, and frailty, not on infarct burden alone. But the large-core trials do dismantle the older assumption that visible infarct extent by itself should exclude a patient — which is exactly the assumption a perfusion gate reintroduces for the less-injured brain.
The precedent is instructive. When the 2015 trials proved anterior-circulation EVT, we did not yet have basilar RCTs — and we did not do nothing for basilar occlusions. We extrapolated from anterior-circulation physiology and registries and opened basilar clots on sound logic. Then BASICS (2021, n=300) was neutral — and yet the field did not abandon a coherent practice, because a single underpowered neutral trial is not the last word. ATTENTION (2022, n=340) and BAOCHE (2022, n=217) subsequently confirmed benefit. The lesson is not that any one neutral or equivocal trial (a BASICS, or a TESLA) settles a question — it is that a biologically coherent practice deserves to stand until better-powered evidence adjudicates it.
The Reframe That Decides the Argument
Strip away the labels and the guideline uses mismatch — a clinical mismatch (a disabling deficit with preserved tissue on ASPECTS). Albers’ strongest version is reasonable: advanced tissue or collateral imaging may better identify the late-window outlier — the small core with little penumbra, or the ≥150 mL core — and where it is available without delaying care, it adds information. The weaker version, the one the correction actually advances, is that this imaging should be mandatory — a gate below which treatment is discouraged. ATLAS shows perfusion is not gating for large core, and DAWN established the principle that a severe deficit out of proportion to the visible core selects responders — though DAWN measured that core with CTP or DWI, so ASPECTS stands in as a surrogate whose calibration is the genuine open question. So the correction’s load-bearing claim narrows to a much more contestable one: that perfusion mismatch is not merely useful but the only acceptable way to select — and on the current evidence, that is more than the data support.
Safety — The Honest Counterweight
The correction is right that harm is not hypothetical at the extremes, and a fair appraisal names the counterweights. TESLA showed more symptomatic hemorrhage with EVT (4.0% vs 1.3%). MR CLEAN-LATE (n=502; overall benefit, adjusted common OR 1.67, 95% CI 1.20–2.32) had significantly more symptomatic ICH with EVT (7% vs 2%; adjusted OR 4.59, 1.49–14.10), and a secondary analysis found a trend toward harm in patients with the smallest penumbra volumes. RESILIENT-EXTEND paired its significant functional-independence gain with a non-significant excess of the worst outcomes, and no benefit above age 68. These signals matter — and they map precisely onto the situations where even a permissive reading should pause: the tiniest penumbra, the ≥150 mL core presenting late, and the older patient. But bounding the edges is not the same as sinking the standard. Across the large-core trials and ATLAS, CT-based selection produced net functional benefit — a result that must be weighed against, not erased by, the hemorrhage and procedural risks that attend thrombectomy in any window.
The Conflict of Interest That Must Be Named
Dr Albers founded and holds equity in iSchemaView (RAPID), and several co-authors report consulting relationships with the same company. Perfusion selection is not vendor-specific — it can be performed on multiple platforms or by MR diffusion — so this is context, not a rebuttal, and the disclosure is complete. But because the correction’s central demand is mandatory perfusion imaging, the relationship is relevant background against which to judge the strength of the evidence on its own terms.
Access, Equity, and the Medicolegal Edge
The correction warns that a Class 1A NCCT-only recommendation could “inadvertently discourage evidence-based practice” and expose clinicians to quality-metric and medicolegal pressure. Fair — but the argument cuts both ways, and harder in the other direction. Most late-presenting patients arrive at primary stroke centers without RAPID. A perfusion mandate could delay transfer or exclude otherwise-treatable patients that TENSION, ANGEL-ASPECT, and ATLAS indicate can benefit — and would encode a two-tier system in which the neuro-emergency you can treat depends on the software your hospital purchased.
The Verdict
- Large core (ASPECTS ≤5): the correction is on the wrong side of ATLAS. CT-based selection is justified to 24 hours — anchored directly by TENSION to 12 hours, and from 12–24 hours by ATLAS plus TESLA’s concordant near-miss — with the honest exceptions of the ≥150 mL core beyond 6 hours and the very elderly.
- Smaller core (ASPECTS 6–10): the clinical mismatch (NIHSS ≥6 + ASPECTS ≥6) is a reasonable pragmatic surrogate for perfusion, and the one pure NCCT/CTA trial — RESILIENT-EXTEND — showed a significant functional-independence benefit; but its uninterpretable ordinal primary and age effect mean the strength of this recommendation should be stated more modestly than for large core.
- Where the correction is right: the ≥150 mL late core, and that NIHSS ≥6 is a looser mismatch than DAWN validated. On those grounds a flat 1A may be a stretch — but IIb is too timid. The defensible level is Class IIa: treat on NCCT + clinical selection; obtain perfusion when available and let it flag the outliers, but do not make it the gatekeeper.
The clinical dilemma resolved: the reason it feels wrong to demand more imaging for the less-injured brain is that it is wrong — once a disabling deficit is present, the small core with a large deficit is the best thrombectomy candidate we have, not the one who needs to clear an extra hurdle.
Late-window LVO (6–24 h), ICA/M1, disabling deficit (NIHSS ≥6), premorbid independence (mRS 0–1) → treat.
ASPECTS ≥6: NCCT + CTA is a reasonable basis to treat; perfusion, if quickly available, adds confidence in the borderline NIHSS 6–8 patient.
ASPECTS 3–5 (large core): treat on NCCT + CTA to 24 h in selected patients — strongest evidence <80 years; weigh mass effect and ≥150 mL extent, the strata where benefit is least certain.
Withholding EVT from an otherwise-eligible patient solely for the absence of perfusion imaging is not evidence-based.
Master Trial Table
| Trial | Yr | Window | Selection / population | N | Result | Bearing on the debate |
|---|---|---|---|---|---|---|
| MR CLEAN / ESCAPE / EXTEND-IA / SWIFT-PRIME / REVASCAT | 2015 | <6–12 h | CTA ± perfusion; small-core anterior LVO | ~1,300 | EVT benefit | Foundation; basis for later extrapolation |
| DAWN | 2018 | 6–24 h | Clinical–core mismatch (NIHSS ≥10) | 206 | mRS benefit | Late window, clinical mismatch |
| DEFUSE 3 | 2018 | 6–16 h | Perfusion mismatch, NIHSS ≥6, core <70 mL | 182 | mRS benefit | Same NIHSS ≥6 the guideline uses + perfusion |
| AURORA (pooled IPD) | 2022 | 6–24 h | 6-trial pool; predominantly mismatch-selected | 505 | acOR 2.54 (1.83–3.54); benefit concentrated in mismatch strata | Albers’ key citation; non-mismatch subgroup small/uncharacterized |
| RESILIENT-EXTEND | 2024 (ISC) | 8–24 h | NCCT + CTA only; ASPECTS 5–10 | 245 | mRS 0–2 25% vs 14% (aOR 2.56, P=.012); ordinal primary “not applicable” (bidirectional); no benefit >68 y | Only pure NCCT/CTA late RCT — signal for it |
| MR CLEAN-LATE | 2023 | 6–24 h | NCCT + CTA collaterals; DAWN/DEFUSE-eligible excluded | 502 | acOR 1.67 (1.20–2.32); sICH 7% vs 2%; harm trend at smallest penumbra | Benefit outside perfusion selection; safety signal at the extreme |
| RESCUE-Japan LIMIT | 2022 | ≤24 h | ASPECTS 3–5 (MRI); large core | 203 | mRS 0–3 benefit | Large-core, positive |
| SELECT-2 | 2023 | ≤24 h | ASPECTS 3–5 or core ≥50 mL | 352 | mRS shift benefit | ASPECTS (NCCT) was a valid entry route |
| ANGEL-ASPECT | 2023 | ≤24 h | ASPECTS 3–5 or core 70–100 mL | 456 | mRS shift benefit | Same; China |
| TENSION | 2023 | ≤12 h | NCCT-predominant (82% CT, 18% MRI) ASPECTS 3–5 | 253 | acOR 2.58 (1.60–4.15); stopped early for efficacy; lower mortality | Cleanest standard-imaging trial — but ≤12 h |
| TESLA | 2024 | ≤24 h | NCCT/ASPECTS 2–5 | 300 | uw-mRS diff 0.63; posterior 0.96 vs 0.975 threshold — near-miss; sICH 4.0% vs 1.3% | Albers’ fulcrum; equivocal, not clearly positive |
| LASTE | 2024 | ≤6.5 h | ASPECTS 0–5; very large core | 333 | Benefit, stopped early | Even huge cores benefit early |
| ATLAS (IPD meta-analysis) | 2026 | ≤24 h | Pooled 6 trials; central adjudication (mixed NCCT/CTP/DWI) | 1,886 | aGenOR 1.63 (1.42–1.88); benefit to core 150 mL, irrespective of time; ≥150 mL late uncertain | Establishes large-core EVT works; weakens the ≤12 h restriction |
| BASICS | 2021 | ≤6 h | Basilar LVO | 300 | Neutral | Precedent: neutral trial didn’t stop practice |
| ATTENTION | 2022 | ≤12 h | Basilar LVO (2:1) | 340 | mRS benefit; ↑ procedural/ICH risk | Precedent: extrapolation later vindicated |
| BAOCHE | 2022 | 6–24 h | Basilar LVO | 217 | mRS benefit | Precedent: extrapolation later vindicated |
Trial sizes and window details reflect the primary publications; central-adjudication figures are from ATLAS (2026).