Every minute counts in the management of a stroke. Yet the diagnostic chain that leads from a symptomatic patient to a therapeutic decision (thrombolysis, mechanical thrombectomy, or conservative treatment) rests on a series of imaging decisions whose complexity is often underestimated. Does this patient genuinely have a large vessel occlusion? Is this cerebral hemorrhage primary, or does it reveal a treatable secondary cause? What is the embolic source in strokes without atrial fibrillation or visible thrombus? Can the emergency CT scan, beyond its role in ruling out hemorrhage, provide risk-stratifying information on functional outcome?
These four questions run through current research in vascular neuroimaging and structure the way care teams organize the initial workup of stroke patients. The accumulation of recent data on clinical triage tools, CT signs of secondary hemorrhage, the contribution of radiomics to risk stratification, and the role of cardiac CT in etiological workup offers a renewed vision of what imaging can contribute, well beyond the simple detection of the lesion.
Identifying Large Vessel Occlusion Before Imaging
The decision to transport a suspected stroke patient directly to a center equipped with a thrombectomy unit rather than to the nearest hospital rests, in the absence of prehospital neuroimaging, on clinical triage tools. These scores, which standardly assess motor function, language, alertness, vision, and sensory deficits, aim to predict the presence of a proximal arterial occlusion, the sole indication for thrombectomy.
A structuring finding emerges: clinical triage tools for large vessel occlusions offer limited added value once the patient has arrived in an emergency department with rapid access to CT angiography. In this high-resource context, CT remains the undisputed arbiter. Their utility nonetheless remains real in two specific situations: prehospital environments where paramedics must decide on the destination center without available imaging, and community hospitals lacking immediate access to CT angiography. In these configurations, coupling a precise triage score with a telemedicine call to a tertiary center would refine patient routing and activate the interventional neuroradiology team before the patient even arrives.
An important limitation to highlight: the majority of occlusions in this cohort involved the M2 segment of the middle cerebral artery (36.9%), a distal location for which all scores are notoriously less accurate. As eligibility criteria for thrombectomy expand to medium and distal vessel occlusions, the question of the sensitivity of triage tools for these locations is becoming an increasingly important clinical issue.
Cerebral Hemorrhage on Non-Contrast CT: When the Image Speaks to the Cause Before the Vascular Workup
When faced with non-traumatic intracranial hemorrhage on non-contrast CT, the clinical reflex is often to attribute the lesion to small vessel disease (chronic arterial hypertension or cerebral amyloid angiopathy) and treat accordingly. But 10 to 20% of primary cerebral hemorrhages are in reality caused by a secondary etiology: arteriovenous malformation, intracranial aneurysm, cerebral venous thrombosis, primary or metastatic tumor, vasculitis, or hemorrhagic transformation of an infarction. These causes require specific etiological interventions, sometimes urgent, sometimes contraindicated if the wrong etiology is assumed, as with anticoagulation which is lifesaving in venous thrombosis but dangerous in most other hemorrhages.
A recent systematic review identified and categorized non-contrast CT signs suggestive of a secondary etiology. Four major semiological domains emerge from this synthesis.
The first concerns the intraparenchymal morphology of the bleeding. A concave, “cashew-shaped” hematoma, juxtacortical and small in size, is nearly pathognomonic of cerebral venous thrombosis. An irregular, “flame-shaped” hematoma may suggest an arteriovenous malformation, aneurysm, or dural fistula. A perilesional edema disproportionate to the hematoma volume points toward an underlying tumor; a perilesional-to-volume ratio above 0.70 proves discriminating between neoplastic and non-neoplastic hemorrhages. Calcifications within or at the periphery of the bleeding suggest an arteriovenous malformation or cavernoma. The fluid-level sign, reflecting sedimentation of cells in a poorly coagulated hematoma, is highly specific for hematological coagulation disorders.
The second domain concerns location. Hemorrhages outside the classic deep supratentorial territories of small vessel disease (cerebral lobes, cerebellum, brainstem) are more often secondary. Multiple or bilateral hemorrhages point toward venous thrombosis, metastatic tumor, vasculitis, or coagulopathy. A hemorrhage in the corpus callosum suggests a primary brain tumor. A location adjacent to venous drainage zones (lateral temporal region, parasagittal region, bilateral thalami) or to typical aneurysmal sites (temporopolar region, paramedian frontal lobe) should trigger an immediate diagnostic alert.
The third domain groups non-hemorrhagic signs: spontaneous hyperdensity of a venous structure (cord sign, dense triangle sign) indicates venous thrombosis with high specificity. The coexistence of ischemic and hemorrhagic lesions points toward posterior reversible encephalopathy syndrome, reversible cerebral vasoconstriction syndrome, or vasculitis. Finally, the absence of small vessel disease markers should, in itself, raise suspicion of a secondary etiology.
The practical value of this semiology is greatest in contexts where MRI and CT angiography are not immediately accessible, and in resource-limited countries that bear the heaviest global burden of cerebral hemorrhages.
Radiomics on Brain CT Angiography: Toward Functional Risk Stratification from the Initial Workup
Beyond detecting the occlusion, a major clinical question arises from the initial CT angiography: what will this patient’s outcome be at discharge and at three months, based on the characteristics of the brain parenchyma at the time of management? This information partly conditions therapeutic decisions, particularly the decision to intervene in the case of large-territory infarction, and shapes communication with families.
Radiomics, the extraction and analysis of high-dimensional quantitative features from medical images, opens a promising avenue for this early stratification. Applied to the middle cerebral artery territories on admission CT angiography of patients with large vessel occlusion who underwent thrombectomy, this approach extracts more than a thousand first-order and texture features from both hemispheres, even before any treatment.
This approach illustrates a major conceptual shift in the use of CT angiography in emergency settings: the examination is no longer solely an anatomical detection tool for the occlusion, but a reservoir of biological information invisible to the human eye, encoded in parenchymal density variations and accessible via algorithms capable of extracting prognostic value from them. The next step consists of integrating these radiomic signatures into clinically usable decision-support tools, deployable in real time within the emergency radiology workflow.
Cardiac CT in the Acute Phase: Unmasking the Embolic Source
Approximately one third of ischemic strokes due to large vessel occlusion remain without identified etiology after standard workup; they are classified as embolic strokes of undetermined source. Among these cases, a significant proportion would in reality be of cardioembolic origin, without documented atrial fibrillation or visible intracardiac thrombus. It is precisely this subgroup that cardiac CT performed in the acute phase allows investigation of.
The value of cardiac CT as a non-invasive complement to transesophageal echocardiography, which is the reference standard but semi-invasive and often impractical in an emergency setting, then comes into question within the initial workup of stroke with large vessel occlusion. Integrating cardiac CT into the first-line imaging protocol allows structured etiological information to be retrieved from the hyperacute phase, rapidly guides secondary anticoagulation decisions, and avoids leaving a significant fraction of severe strokes unexplained.
Toward Stroke Imaging That Reinvents Itself at Every Level of the Care Chain
Recent data paint a picture of emergency neurological imaging in full transformation. From prehospital triage to post-thrombectomy etiological workup, every link in the diagnostic chain is now being challenged by new approaches, whether clinical, semiological, quantitative, or structural, that make it possible to extract more information from the same examinations.
The overarching lesson is one of complementarity between tools. The triage score is useful where CT cannot yet intervene. The CT semiology of cerebral hemorrhage guides the decision to perform supplementary imaging when MRI is unavailable. Radiomics exploits what the eye cannot see in parenchymal densities. Cardiac CT reveals embolic mechanisms that conventional examinations miss. None of these tools functions alone; it is their reasoned articulation within a coherent protocol that creates clinical value.
The prospects for research and industry are numerous. Validation of triage scores under real prehospital conditions, coupled with telemedicine models, remains a priority for healthcare systems covering large geographical areas. The automation of recognition of CT signs suggestive of secondary hemorrhage, a domain in which detection algorithms could assist the emergency physician or a radiologist with limited neuroradiology experience, represents a high-impact industrial project, particularly for resource-limited environments. The integration of radiomic signatures into real-time risk stratification tools, deployable directly within the radiologist’s reading console, is now technically accessible and clinically relevant. Finally, the definition of cardiac imaging protocols integrated into the initial workup of severe strokes, standardized and validated on international multicenter cohorts, is the next step to transform these exploratory data into practice recommendations.
Sources
Desmeules F, Emond M, Nadeau A, et al. Accuracy of Published Screening Tools for Large Vessel Occlusion in Patients With Suspected Acute Ischemic Stroke: A Prospective Cohort Study. Ann Emerg Med. 2026 Jan;87(1):15-24. doi: 10.1016/j.annemergmed.2025.07.030. PMID: 40974366. https://pubmed.ncbi.nlm.nih.gov/40974366/
Mostafa K, Wünsche C, Krutmann S, et al. Cardiac CT in Large Vessel Occlusion Stroke for the Evaluation of Non-Thrombotic and Non-Atrial-Fibrillation-Related Embolic Causes. Neurol Int. 2025 Feb 7;17(2):25. doi: 10.3390/neurolint17020025. PMID: 39997656. https://pubmed.ncbi.nlm.nih.gov/39997656/
Avery EW, Behland J, Mak A, et al. Dataset on acute stroke risk stratification from CT angiographic radiomics. Data Brief. 2022 Aug 14;44:108542. doi: 10.1016/j.dib.2022.108542. PMID: 36060820. https://pubmed.ncbi.nlm.nih.gov/36060820/
Pensato U, Rapillo CM, Mazzacane F, et al. Non-contrast CT findings suggestive of secondary intracerebral haemorrhage. Eur Stroke J. 2026 Jan 1;11(1):aakaf010. doi: 10.1093/esj/aakaf010. PMID: 41614518. https://pubmed.ncbi.nlm.nih.gov/41614518/
