Skip to content

Biomarkers and imaging markers

TL;DR — Stroke already uses biomarkers, but they are mainly imaging phenotypes: hemorrhage on CT, occlusion on angiography, diffusion lesion, perfusion mismatch, collateral grade, hematoma volume, microbleeds, and tract integrity. A PubMed update through 2026-08-30 found no blood test externally validated to replace urgent brain imaging for distinguishing ischemic from hemorrhagic stroke; a 141-study review found statistical signals across 136 candidates but “few” with meaningful clinical value (Hasan 2012, PMID 22320313). A 2026 prospective acute-vertigo cohort illustrates the boundary: plasma neurofilament light had 84.2% sensitivity but only 62.4% specificity for posterior-circulation stroke in 102 patients, supporting further triage research rather than imaging replacement (Haidegger 2026, PMID 41961296). Imaging biomarkers can select treatment: mismatch-selected unknown-onset thrombolysis increased mRS 0–1 from 39% to 47% (adjusted OR 1.49, 95% CI 1.10–2.03) while increasing symptomatic ICH from <1% to 3% (Thomalla 2020, PMID 33176180). Prognostic biomarkers often recapitulate severity, age, infarct size, or inflammation without adding enough validated discrimination to change care. A useful biomarker must improve a specified decision in a new population—not merely achieve a significant association in the development cohort.

A decision-based taxonomy

Intended use Example question Required comparison
Diagnostic Is this stroke, and ischemic or hemorrhagic? Standard clinical assessment plus urgent imaging
Etiologic Is the mechanism cardioembolic, atherosclerotic, or small-vessel? Full mechanism workup and adjudication
Treatment selection Is tissue salvageable or hemorrhage likely? Existing eligibility criteria and net clinical benefit
Prognostic What disability, death, or recurrence risk? Validated clinical score and imaging baseline
Monitoring Has reperfusion, edema, bleeding, or recovery changed? Repeated standard measurement
Predictive Does treatment effect differ by marker? Randomized marker-by-treatment interaction

Prognostic association is not treatment prediction. A marker can identify poor outcome yet fail to identify who benefits from an intervention.

Hyperacute structural and vascular imaging

Marker Captures Validated role Limitation
Non-contrast CT hemorrhage Acute blood and mass effect First treatment branch Early ischemia may be normal
ASPECTS Extent of early MCA-territory ischemic change Rapid standardized estimate Regional weighting and reader dependence
CTA occlusion site Treatable arterial target Thrombectomy selection Single phase may undergrade delayed collaterals
CTA collateral grade Alternative downstream filling Prognosis and tissue-tempo context Acquisition/scale dependent
DWI/ADC lesion Cytotoxic edema Sensitive acute lesion localization 6.8% pooled DWI-negative stroke
Perfusion core/mismatch Modelled low flow and at-risk tissue Extended-window selection Software/threshold dependence
Susceptibility blood products Hemorrhage, thrombus sign, microbleeds Hemorrhage and small-vessel phenotype Sequence-dependent sensitivity

DWI-negative acute ischemic stroke occurred in 6.8% (95% CI 4.9–9.3) across 3,236 patients, with posterior-circulation stroke odds 5.1 times higher than anterior circulation (Edlow 2017, PMID 28615423). CT perfusion diagnostic sensitivity was 80% (72–86) and specificity 95% (86–98) across 1,107 suspected strokes; small lacunes caused most false negatives (Biesbroek 2013, PMID 23736122).

Tissue clocks and treatment selection

Imaging mismatch converts continuous biology into an eligibility rule. DAWN used clinical deficit–core mismatch up to 24 hours, while DEFUSE 3 used perfusion-defined mismatch up to 16 hours (Nogueira 2018, PMID 29129157; Albers 2018, PMID 29364767).

In four unknown-onset thrombolysis trials (843 participants), advanced-imaging selection yielded:

Outcome Alteplase Control Adjusted effect
mRS 0–1 at 90 days 47% 39% OR 1.49 (1.10–2.03)
Independent outcome — — OR 1.50 (1.06–2.12)
Symptomatic ICH 3% <1% OR 5.58 (1.22–25.50)
Death 6% 3% OR 2.06 (1.03–4.09)

Source: Thomalla 2020, PMID 33176180. The biomarker is useful because randomized data connect it to net treatment benefit, not because mismatch independently predicts outcome.

Collaterals and reperfusion markers

Good collaterals associate with slower infarct growth, successful reperfusion, less hemorrhagic transformation, and better outcome, but treatment decisions based on collateral grade alone lack comparable randomized validation (Lee 2023, PMID 35687300). Across 24 studies and 2,239 thrombectomy patients, good collaterals associated with successful reperfusion RR 1.28 (95% CI 1.17–1.40) and recanalization RR 1.23 (1.06–1.42) (Leng 2016, PMID 26579719).

Post-treatment marker Meaning Failure mode
eTICI reperfusion grade Angiographic downstream reperfusion Does not prove microvascular or tissue reperfusion
Final infarct volume Tissue injury after treatment Affected by timing, edema, segmentation, and survival
Early neurological change Composite biological/clinical response Sedation, BP, glucose, seizure, and hemorrhage confound
BBB permeability/contrast staining Barrier injury Can resemble hemorrhage

Blood markers: diagnosis

A 2012 systematic review identified 141 studies of 136 blood candidates. CRP, P-selectin, and homocysteine differed between ischemic stroke and healthy controls; GFAP differed between hemorrhagic and ischemic stroke by a pooled mean 224.58 ng/L (95% CI 25.84–423.32). Heterogeneous timing, assays, controls, and spectrum limited clinical translation (Hasan 2012, PMID 22320313).

Candidate family Biological rationale Present limitation
GFAP/S100B Astroglial injury; hemorrhage may release rapidly Timing and overlap prevent imaging replacement
Neurofilament light Axonal injury burden Not stroke-specific
D-dimer/fibrin markers Thrombosis and clot burden Elevated in cancer, infection, age, venous thrombosis
CRP/cytokines Systemic/neuroinflammation Low specificity and reverse causation
microRNAs/transcripts Cell-state signatures Platform, normalization, and replication problems
Metabolomic/proteomic panels Multidimensional classification Overfitting and implementation complexity

Presenting symptoms and signs alone also have insufficient discrimination to substitute for imaging (McDermott 2025, PMID 40759192).

Blood markers: deterioration and hemorrhagic transformation

Early neurological deterioration occurred in 11.9% using stricter and 18.6% using looser definitions in a review of 82 studies. Associated admission differences included glucose +0.90 mmol/L, hs-CRP +3.79 mg/L, leukocytes +0.54×10^9/L, and fibrinogen +0.32 g/L; incremental value beyond clinical/imaging models remained unproven (Martin 2018, PMID 30517919).

MMP-9 is the most studied hemorrhagic-transformation candidate:

  • Seven pooled studies produced sensitivity 85% (95% CI 75–91), specificity 79% (67–87), and AUC 0.89, but heterogeneity exceeded 50% and cutoffs differed (Wang 2018, PMID 29598905).
  • A later review reported pooled diagnostic OR 29.57 (17.75–49.27) for MMP-9; c-Fn had DOR 299 but CI 20.5–4,366.7, illustrating instability from small evidence sets (Krishnamoorthy 2022, PMID 34569521).
  • Imaging markers of ischemic extent and BBB permeability remain operationally closer to the injury process and immediately available, but also lack a universally sufficient prediction rule (Hong 2022, PMID 35026765).

Large odds ratios do not establish clinical utility when thresholds, prevalence, assay turnaround, calibration, and treatment consequences are unresolved.

Broader hemorrhagic-transformation reviews converge on the same limitation: animal–human differences, heterogeneous radiological definitions, and overlapping inflammatory, endothelial, and coagulation pathways prevent a single candidate from becoming a treatment gate (Jickling 2014, PMID 24281743; Lu 2018, PMID 28726570; Liu 2022, PMID 33125600; Qiu 2021, PMID 34248961; Thomas 2021, PMID 34912581; Kovács 2023, PMID 37762370). BBB-focused imaging may ultimately complement rather than replace molecular panels (Candelario-Jalil 2022, PMID 35387495).

General prognostic markers

Marker Reported association Why it is not yet a routine decision test
Copeptin Mortality OR 4.16 (2.77–6.25); poor outcome OR 2.56 (1.97–3.32), 1,976 patients Stress marker; external calibration and action threshold unclear
Admission glucose Poor outcome and hemorrhage association Confounded by diabetes and stress; treatment effect differs from association
Uric acid Outcome associations in meta-analysis Nonlinear/confounded biology and assay timing
Protein Z/coagulation factors Case-control associations Mechanistic heterogeneity and no standardized threshold
NLR Hemorrhagic-transformation DOR 5.04 (2.90–8.75) Non-specific inflammation and variable timing

Sources: Xu 2017, PMID 27904159; Hasan 2012, PMID 22320313; Zhang 2021, PMID 34625360; Słomka 2020, PMID 32369852; Krishnamoorthy 2022, PMID 34569521.

Recovery biomarkers

Recovery-marker studies use corticospinal-tract lesion load, diffusion tensor integrity, motor-evoked potentials, resting/task fMRI, EEG, and structural lesion topology. Among 71 motor-recovery studies, only 21 met a ≥80% methodological-quality threshold; common failures were absent cross-validation, small samples, and failure to use a minimal clinically important difference (Kim 2017, PMID 27503908).

Modality Candidate signal Translational constraint
Structural MRI Lesion site/volume and tract overlap Collinearity with clinical severity
DTI Corticospinal fractional anisotropy/asymmetry Scanner and pipeline dependence
TMS Presence of motor-evoked potential Limited availability and contraindications
fMRI Network activation/connectivity Motion, task performance, analytic flexibility
EEG/QEEG Slowing, asymmetry, connectivity State and artifact dependence

QEEG meta-analysis included only 482 participants: higher delta–alpha ratio correlated with worse mRS at r=0.26 (95% CI 0.21–0.31), and delta–theta–alpha–beta ratio at r=0.32 (0.26–0.39) (Sood 2024, PMID 39357611). These modest associations do not yet prescribe rehabilitation type or dose.

EEG/MEG sensorimotor reviews identify methodological heterogeneity (Tedesco Triccas 2019, PMID 30118725), while aphasia electrophysiology remains similarly exploratory (Arheix-Parras 2023, PMID 36749552). Recovery-roundtable standards were created partly because incompatible time points and outcomes prevented aggregation (Bernhardt 2017, PMID 28697708).

Validation requirements

  1. Pre-specify the clinical decision and target population.
  2. Use an assay or imaging pipeline reproducible across sites.
  3. Compare incremental discrimination, calibration, and decision benefit against standard predictors.
  4. Validate externally across age, sex, ancestry, mechanism, severity, and care setting.
  5. Demonstrate turnaround compatible with the decision window.
  6. For a predictive marker, test the marker-by-treatment interaction in randomized data.
  7. Quantify missingness, indeterminate results, and harms of false classifications.

Machine-learning accuracy without calibration and external validation is not a clinical biomarker. A locked model, defined threshold, and prospective workflow test are minimum steps.

Open questions

  • Can a rapid blood panel add enough information to imaging to safely treat before complete transfer, without missing hemorrhage? (Hasan 2012, PMID 22320313)
  • Which BBB marker improves hemorrhagic-transformation prediction beyond core volume, glucose, age, and treatment delay? (Krishnamoorthy 2022, PMID 34569521; Hong 2022, PMID 35026765)
  • Does collateral grade modify thrombectomy treatment effect, rather than only prognosis? (Lee 2023, PMID 35687300)
  • Can recovery biomarkers prospectively assign a rehabilitation intervention or dose that improves outcome? (Kim 2017, PMID 27503908; Bernhardt 2017, PMID 28697708)
  • How much apparent biomarker performance disappears under multicenter assay harmonization and locked external validation? (Martin 2018, PMID 30517919)

References

  1. Hasan N, et al. Blood biomarkers for acute stroke in humans: systematic review. Br J Clin Pharmacol. 2012. PMID 22320313
  2. McDermott CM, et al. Discrimination of ischemic versus hemorrhagic stroke by symptoms or signs. J Stroke Cerebrovasc Dis. 2025. PMID 40759192
  3. Edlow BL, et al. DWI-negative acute ischemic stroke: meta-analysis. Neurology. 2017. PMID 28615423
  4. Biesbroek JM, et al. CT perfusion diagnostic accuracy for acute ischemic stroke. Cerebrovasc Dis. 2013. PMID 23736122
  5. Thomalla G, et al. Imaging-guided alteplase for unknown-onset stroke: individual-patient meta-analysis. Lancet. 2020. PMID 33176180
  6. Nogueira RG, et al. Thrombectomy 6 to 24 hours after stroke. N Engl J Med. 2018. PMID 29129157
  7. Albers GW, et al. Thrombectomy 6 to 16 hours with perfusion selection. N Engl J Med. 2018. PMID 29364767
  8. Lee JS, et al. Collateral status and outcomes after thrombectomy. Transl Stroke Res. 2023. PMID 35687300
  9. Leng X, et al. Collateral status and successful revascularization: meta-analysis. Cerebrovasc Dis. 2016. PMID 26579719
  10. Martin AJ, et al. Molecular biomarkers of early neurological deterioration. Cerebrovasc Dis. 2018. PMID 30517919
  11. Krishnamoorthy S, et al. Biomarkers predicting hemorrhagic transformation. Cerebrovasc Dis. 2022. PMID 34569521
  12. Wang L, et al. Serum MMP-9 for predicting hemorrhagic transformation. J Stroke Cerebrovasc Dis. 2018. PMID 29598905
  13. Hong L, et al. Neuroimaging prediction of hemorrhagic transformation. Cerebrovasc Dis. 2022. PMID 35026765
  14. Jickling GC, et al. Hemorrhagic transformation after ischemic stroke in animals and humans. J Cereb Blood Flow Metab. 2014. PMID 24281743
  15. Liu C, et al. Hemorrhagic transformation after tissue plasminogen activator. Cell Mol Neurobiol. 2022. PMID 33125600
  16. Kovács KB, et al. Hemorrhagic transformation of ischemic strokes. Int J Mol Sci. 2023. PMID 37762370
  17. Lu G, et al. Potential biomarkers predicting hemorrhagic transformation. Int J Neurosci. 2018. PMID 28726570
  18. Thomas SE, et al. Risk factors and predictors for hemorrhagic transformation. Int J Vasc Med. 2021. PMID 34912581
  19. Xu Q, et al. Copeptin for prognosis after ischemic stroke and TIA: meta-analysis. Hypertens Res. 2017. PMID 27904159
  20. Zhang M, et al. Uric acid and acute ischemic stroke prognosis: meta-analysis. Nutr Metab Cardiovasc Dis. 2021. PMID 34625360
  21. Słomka A, et al. Plasma protein Z in ischemic stroke: meta-analysis. Thromb Haemost. 2020. PMID 32369852
  22. Kim B, Winstein C. Neurological biomarkers predicting poststroke motor recovery. Neurorehabil Neural Repair. 2017. PMID 27503908
  23. Sood I, et al. Quantitative EEG and post-stroke disability: meta-analysis. J Stroke Cerebrovasc Dis. 2024. PMID 39357611
  24. Tedesco Triccas L, et al. EEG/MEG and upper-limb impairment after stroke. J Neurosci Methods. 2019. PMID 30118725
  25. Arheix-Parras S, et al. Electrophysiological changes in post-stroke aphasia. Brain Topogr. 2023. PMID 36749552
  26. Bernhardt J, et al. Standards in stroke recovery research. Int J Stroke. 2017. PMID 28697708
  27. Candelario-Jalil E, et al. Neuroinflammation, BBB dysfunction, and imaging. Stroke. 2022. PMID 35387495
  28. Qiu YM, et al. Immune cells in BBB disruption after ischemic stroke. Front Immunol. 2021. PMID 34248961
  29. Haidegger M, et al. Plasma neurofilament light chain and glial fibrillary acidic protein in the differential diagnosis of acute vertigo in the emergency department. J Neurol. 2026;273. PMID 41961296