Priorities for Advancements in Neuroimaging in the Diagnostic Workup of Acute Stroke

Abstract

The Stroke Treatment Academy Industry Roundtable (STAIR) XII included a workshop to discuss the priorities for advancements in neuroimaging in the diagnostic workup of acute ischemic stroke. The workshop brought together representatives from academia, industry, and government. The participants identified ten critical areas of priority for the advancement of acute stroke imaging. These include enhancing imaging capabilities at primary and comprehensive stroke centers, refining the analysis and characterization of clots, establishing imaging criteria that can predict the response to reperfusion, optimizing the Thrombolysis in Cerebral Infarction (TICI) scale, predicting first-pass reperfusion outcomes, improving imaging techniques post reperfusion therapy, detecting early ischemia on non-contrast computed tomography, enhancing cone beam computed tomography, advancing mobile stroke units, and leveraging high-resolution vessel wall imaging to gain deeper insights into pathology. Imaging in acute ischemic stroke treatment has advanced significantly, but important challenges remain that need to be addressed. A combined effort from academic investigators, industry, and regulators is needed to improve imaging technologies and, ultimately, patient outcomes.

Introduction

During the 12^th Stroke Treatment Academic Industry Roundtable (STAIR XII), international experts from academia, industry and the US government gathered to share their knowledge and seek consensus on strategies intended to surmount impediments in stroke research. This article focuses on ten priorities identified for advancing neuroimaging in the diagnostic workup of acute ischemic stroke (Tables 1 & 2).

Table 1.

Research priorities in imaging and systems of care.

	Current Status	Recommendation
1. Imaging Capabilities at Primary and Comprehensive Stroke Centers	There is no consensus of the optimal imaging protocol for primary and comprehensive stroke centers.	Multi-modal imaging is desirable for primary stroke centers and should be a requirement for comprehensive stroke centers.
2. Clot Analysis and Characterization	There is no platform to determine clot characteristics with non-invasive imaging.	The development of novel post-acquisition processing of noninvasive imaging could potentially determine clot characteristics and improve the effectiveness of EVT.
3. Imaging criteria that predict response to reperfusion	Recent trials have shown the benefit of reperfusion in patients with large core defined as ASPECTS ≤ 5 or volume between 50–100 mL.	Further studies are needed to determine the optimal combination of factors to consider in treatment decision-making for patients with large ischemic core.
4. TICI Score in the Determination of Effective Reperfusion	TICI scores are subjective and do not always translate into tissue reperfused.	Volumetric maps of reperfusion should be generated and validated as potential biomarkers of reperfused brain.
5. First Pass Effect Prediction	There is growing evidence that FPE is a good metric to determine the efficacy of endovascular recanalization.	AI-derived algorithms that encompass patient derived data, clot characteristics and angio-architecture of the target vessel could potentially improve the FPE by recommending specific devices and/or techniques.
6. Imaging Post Reperfusion	The determination of BBB disruption has a low accuracy and is based on the extravasation of contrast into the parenchyma. Similarly, there is no reliable way to quantify cerebral edema.	Newer imaging techniques could potentially identify BBB disruption and characterize cerebral edema. This could guide post-EVT management and be used as a biomarker of neuroprotective therapies.

Table 2.

Optimization of imaging technologies.

	Current Status	Recommendation
7. Detection of early ischemia on Non-Contrast Computed Tomography (NCCT)	NCCT is not accurate in detecting acute ischemic changes. Moreover, its reproducibility is not optimal.	More accurate volumetric and objective measurements of acute ischemia detection should be developed. Continued education of neuroscience trainees to learn how to interpret acute ischemic change is important.
8. Utility of Cone Beam Compute Tomography (CBCT)	CBCT does not accurately determine ASPECTs.	New CBCTs should achieve better grey/white matter differentiation for early detection of acute ischemic changes.
9. Imaging in Mobile Stroke Units (MSU)	MSU have shortened time to thrombolytic with a good safety profile.	CT miniaturization would allow its use in smaller ambulances without compromising image resolution.
10. Vessel Wall Imaging (VWI) to Understand Pathology	It has been recognized as a useful adjuvant in stroke diagnosis and for treatment decisions. However, VWI is not routinely used in clinical practice.	Artificial Intelligence-guided VWI protocols can shorten acquisition times and improve workflows to improve its generalized adoption.

RESEARCH PRIORITIES IN IMAGING AND SYSTEMS OF CARE

1. Imaging capabilities at primary and comprehensive stroke centers

In routine acute stroke practice, critical information for decision-making is obtained from imaging. Hemorrhage is differentiated from ischemia using simple anatomic imaging with non-contrast brain computed tomography (NCCT) or fast protocol magnetic resonance imaging (MRI). Treatment with intravenous thrombolysis may be initiated based on this imaging only. Endovascular therapy (EVT) requires a target arterial occlusive lesion, and this is demonstrated using non-invasive computed tomography angiography (CTA), magnetic resonance angiography (MRA) or, in selected cases, flat-panel dynamic angiography prior to proceeding to arterial access. Choices of how stroke imaging is implemented in clinical routine vary across the world. The ability to easily acquire CTA and computed tomography perfusion (CTP) imaging, also known as multi-modal stroke imaging, on contemporary scanners has made it feasible for most hospitals to perform the initial triage of stroke patients in the hyper-acute stroke setting. Post-processing advancements have facilitated the rapid interpretation of multi-modal imaging of the brain. Several academic and commercial software solutions enable automated large vessel occlusion (LVO) detection and generation within minutes of CTP maps that outline the estimated ischemic core and critically hypoperfused tissue.¹ While the use of perfusion imaging is not required by guidelines for thrombolytic treatment decisions within 4.5h and thrombectomy decisions within 6h, the additional diagnostic and prognostic information can be helpful to support clinical decision making.² For instance, when there is uncertainty regarding the diagnosis of stroke, the presence (or absence) of a perfusion deficit that correlates with the clinical symptoms can offer informative evidence to confirm (or exclude) the diagnosis. Routine use of perfusion imaging also creates greater familiarity with the imaging protocol, which expedites stroke workup, reduces technical errors in image acquisition, and improves image interpretation. However, not all the imaging protocols use the same scan parameters or post-acquisition processing metrics. The estimation of core and penumbra among different academic and commercially available software may vary. However, when specific thresholds and post-processing methods are used to compare different post-acquisition software, a high agreement could be reached. After controlling for these confounding factors, Pisani et al. showed substantial agreement between perfusion parametric maps of three commonly used commercial software packages in a cohort of 242 patients.³ However, to decide if a patient who presents with stroke symptoms within 4.5 hours of last known well is eligible for intravenous thrombolysis, a NCCT to rule out intracerebral hemorrhage and assess the severity of early ischemic injury suffices.⁴ Routine neurovascular imaging with CTA or MRA is advised to determine eligibility for EVT in the early (<6 hours) and late (6–24 hour) time-windows, and provide information on potential causes of stroke.^2,5–7 Current guidelines recommend the acquisition of imaging to determine the presence of penumbra in patients who present in the late time window and who may benefit from EVT. However, these recommendations may change based on the results of recently completed and ongoing ‘large core’ studies.^8–10

The routine use of multi-modal imaging offers additional advantages in stroke research based on a survey of the workshop participants. For example, at comprehensive stroke centers the routine use of multi-modal imaging is recommended to facilitate endovascular stroke research aimed at refining EVT eligibility criteria (e.g., patients with medium vessel occlusions, or low National Institutes of Health Stroke Scale (NIHSS) scores). Similarly, the recently completed TIMELESS study used multi-modal imaging to select patients for intravenous Tenecteplase at both primary and comprehensive stroke centers in the late time-window.¹¹ The low proportion of patients enrolled in TIMELESS at primary stroke centers (5% of study population), highlights the need to increase the number of primary stroke centers that are familiar with multi-modal imaging and that have the infrastructure to conduct clinical stroke trials. Multimodal imaging at primary stroke centers may also benefit trials of brain cytoprotective therapy. Patients transferred from primary to comprehensive stroke centers for EVT are likely ideal candidates for cytoprotective trials. Multimodal imaging may help identify those transfer patients who are most at risk of rapid expansion of their ischemic core and are, therefore, most likely to benefit from cytoprotection.¹²

2. Clot analysis and characterization

Thrombus characterization on pre-EVT imaging has been suggested as a prognostic marker of case complexity, first-pass effect (FPE), and clinical outcome.¹³ Thrombi vary in composition and morphology, resulting in a wide range of thrombus sizes, shapes, cohesion, permeability, and textures. Such thrombus characteristics might be used to guide EVT decisions, device selection, and enable further improvement of procedural and functional outcomes.

Radiomics is a method that aims to quantify the phenotypic characteristics of medical imaging using automated algorithms. Image data is processed by many automatically extracted data-characterization algorithms, referred to as radiomic features (RFs). Radiomics was pioneered in oncology for tumor phenotyping and, more recently, has been applied to stroke imaging.¹⁴ Standard-of-care images such as NCCTs and CTAs, can be transformed with radiomics into quantitative image-based data to enable bioinformatics and AI analyses. This requires segmentation of the NCCT and CTA to determine the thrombi boundaries before data extraction.¹⁵ Thrombus radiomics have been used to predict successful reperfusion, the number of EVT passes, and functional outcomes.¹⁶ A recent analysis of the MR CLEAN registry performed manual segmentations and measurements of thrombi using a 3-dimensional imaging software.¹⁷ Larger volume thrombi were associated with a lower probability of functional independence defined as a modified Rankin Score (mRS) ≤ 2 (OR 0.78, 95% CI 0.64–0.96) and a higher number of retrieval attempts (OR 0.16, 95% CI 0.04–0.28]).¹⁸ A recent study by Santo et al. identified RFs correlated with micro-CT imaging and histopathological samples.¹⁹ RFs computed from NCCT and CTA demonstrated significant association with red blood cells and fibrin-platelet components. Phenotyping clot composition by RFs could potentially guide treatment decisions such as the choice of EVT device or choice of thrombolytic agent and could determine stroke etiology. The STAIR workgroup encourages further research into the use of RFs and other imaging modalities for detailed clot characterization.

3. Imaging criteria that predict response to reperfusion

Early trials of thrombectomy failed to demonstrate a clinical benefit. An important contributing factor was the absence of imaging selection criteria to enrich the trial population with patients who were most likely to benefit from treatment. For example, post hoc analysis of the IMS-3 trial showed that a positive result favoring thrombectomy would likely have been observed if the trial had been limited to patients with evidence of a LVO on baseline imaging.²⁰ Subsequent trials that took this approach and limited inclusion to patients with an LVO showed a substantial benefit from EVT.^21–25 As had previously been observed in trials of intravenous thrombolytics, the treatment effect of EVT diminishes with longer onset-to-treatment times. More recently, trials with ‘penumbral’ selection criteria, have demonstrated that patients with evidence of salvageable tissue on CT or MR imaging benefit from EVT and intravenous thrombolytics even in the late time-window.^26,27 Taken together, these trials have demonstrated the power of vascular and perfusion imaging to identify patients who are likely to benefit from reperfusion therapy. Specifically, patients whose baseline imaging indicates the presence of a LVO involving the internal carotid artery (ICA) or middle cerebral artery (MCA)-M1 segment, a small ischemic core, and a substantial territory of salvageable brain tissue. Patients with these characteristics are more likely to benefit from reperfusion therapy. These criteria have been endorsed in international guidelines.⁷ What remains unanswered is which patients who do not meet the penumbral selection criteria of these trials are nevertheless likely to benefit from reperfusion. Recently published trials of EVT in patients with large cores have tried to address this issue.^28–30 These trials suggest that patients with low Alberta Stroke Program Early CT Score (ASPECTS, ranging from 3 to 5) and/or large cores on CTP or MRI (≥ 50 mL) may benefit from reperfusion. The number needed to treat for benefit, however, is larger than when stricter penumbral selection criteria are applied. In the SELECT2 trial there was no clear upper limit to the core volume associated with thrombectomy benefit in ordinal analysis of mRS.¹⁰ However, the proportion of patients achieving mRS 0–2 or 0–3 was low in individuals with very large core volumes. The core volume remained strongly prognostic but likely needs to be considered alongside other factors including core location, patient co-morbidities and frailty, and patient preferences around acceptable levels of disability. The core and penumbral threshold volumes that determine if a patient is likely to benefit from EVT remain unknown, and most likely would have to be tailored for each patient.

Workshop participants recognized the importance of understanding the potential combinations of factors that may cause EVT to be futile or even detrimental. A recent sub-analysis of RESCUE-Japan LIMIT compared outcomes of patients with ASPECTS ≤ 3 versus 4 and 5. EVT was not associated with improved functional outcome at 90 days in patient with ASPECTS ≤ 3. Moreover, this group had a higher incidence of symptomatic ICH.³¹ The median ischemic core in the ASPECTS ≤ 3 group was 126 cc, versus 89 cc in the ASPECTS 4–5 group (p<0.001). It should be noted that RESCUE-Japan LIMIT was primarily an MRI selection study in which ASPECTS was assessed on diffusion weighted imaging (DWI) in 86.1% of the cohort. This approach yields, on average, ASPECTS scores that are 1 point lower than ASPECTS measured by NCCT.³²

Individual patient data meta-analysis may further provide data on the potential benefit of EVT in patients with very large cores and/or without the presence of a salvageable penumbra on baseline imaging.

4. Thrombolysis in cerebral infarction (TICI) score in the determination of effective reperfusion.

The thrombolysis in cerebral infarction (TICI) scale is a widely used scoring system to evaluate the degree of reperfusion achieved after mechanical thrombectomy (MT). The TICI scale was originally proposed in a position statement that attempted to standardize clinical trial design and reporting for intra-arterial therapy.³³ The TICI grading system is divided in three grades, 0 corresponds to no perfusion and 3 to complete perfusion, with antegrade flow into the bed distal to the obstruction.³³ Grade 2 can be divided into partial filling with less than two thirds of the entire vascular territory (2a) or complete filling of all the expected vascular territory, but the filling is slower than normal (2b).³⁴ Studies have demonstrated that a detailed six-step grading scale is more accurate in determining clinical outcomes, than the standard TICI grading system.³⁵ The inclusion of a TICI 2c to label patients with near-complete reperfusion, except for slow flow in one or two distal cortical vessels or the presence of minor distal emboli provides a more granular assessment of reperfusion than the standard TICI scoring system.³⁶ The HERMES group core laboratory introduced an expanded TICI (eTICI), encompassing all the various thresholds used to define reperfusion after EVT.³⁷ The eTICI system further refines grade 2 into distinct percentages of perfusion. Specifically, eTICI 2a denotes reperfusion in less than half or 1–49% of the affected territory; eTICI 2b50 indicated 50–66%reperfusion; eTICI 2b67 represents 67–89% reperfusion, eTICI 2c is equivalent to TICI 2C or 90–99%reperfusion; and eTICI 3 denotes complete or 100% reperfusion, akin to TICI 3. Despite these improvements in defining reperfusion, TICI-based systems of determining effective reperfusion are highly subjective and may not reflect restoration of blood flow in microcirculatory vessels. Moreover, TICI assessments by visual inspection are prone to error, as they may be affected by the experience level of the rater, operator bias, and field of view. TICI scores are generally overestimated by operators during EVT compared to core-lab raters.³⁸ The inclusion of finer scales and artificial intelligence (AI)-based automated protocols can potentially provide a better assessment of reperfusion and may be a better prognostic tool than coarser scales. A study by Prasetya et al. used a semiautomated platform for the segmentation of the downstream vascular territory of the occluded vessel.³⁹ qTICI was defined as the percentage of reperfused area in the target downstream territory. The determination of reperfusion with qTICI was comparable with eTICI and performed similarly in predicting favorable outcome. A study by Su et al. used convolutional neural networks to generate a fully automatic and quantitative perfusion-based TICI score. This autoTICI performed on par with human experts.⁴⁰ On the MR CLEAN registry, there was a statistically significant association between autoTICI and eTICI, and both accurately predicted functional outcome.⁴⁰

The STAIR workgroup recommends the development and implementation of automated perfusion scores in assessing flow to the downstream target territory. Automating these scores will eliminate subjectivity, improve standardization, and facilitate comparison among studies. Granular automated data would ultimately be used for the estimation of reperfusion of eloquent territories.

5. First-pass effect prediction

The first-pass effect (FPE) concept entails achieving near complete or complete revascularization of the occluded large vessel and its downstream territory (mTICI 2c/3) through a single revascularization attempt without the need for rescue therapy.⁴¹ FPE is associated with better clinical outcomes, lower mortality, and fewer procedural adverse events.⁴¹ As a result, FPE has been proposed as a potential benchmark to assess the technical efficacy of EVT techniques and devices and as a potential surrogate measure for their clinical efficacy. The analysis of a cohort of 930 patients demonstrated that FPE could be achieved in 40.5% of patients.⁴² This study reported two variables as independent predictors of FPE: non-ICA occlusion and the use of a balloon-guide catheter for EVT. Achievement of FPE may be a function of three interrelated factors: patient-related variables, occlusion characteristics, and procedural factors. Patient-related predictors include age and stroke etiology.⁴³ Occlusion-related predictors are associated with the occlusion’s location, the clot’s characteristics, and the angio-architecture of the target occlusion. Device and technique-related variables include using balloon-guide catheters,^43,44 the device length in case of stentretrievers,⁴⁵ and bore size of the aspiration catheter in case of contact aspiration devices.⁴⁶ The analysis of the MR CLEAN Registry showed that history of hyperlipidemia (OR 1.05; 95% CI, 1.01–1.10), MCA occlusion versus intracranial ICA occlusion (OR, 1.11; 95% CI, 1.06–1.16), and aspiration versus stent thrombectomy (OR, 1.07; 95% CI, 1.03–1.11) were associated with FPE.⁴⁷ Neurointerventionalist experience increased the likelihood of FPR (OR, 1.03 per 50 patients previously treated; 95% CI, 1.01–1.06). Therefore, the technical acumen of the neurointerventionalist is also an important factor in FPE.

Whether FPE can be used as a valid surrogate endpoint in EVT trials remains to be determined. Although, the latest advancements in AI have opened opportunities for the integration of machine learning in determining FPE,¹⁶ considerable validation effort remains to be done. Notably, the assessment of clot characteristics through NCCT and CTA has been leveraged to predict the ease of clot extraction with variable success.^48,49 Other AI-based models that account for patient-specific characteristics, clot information derived from non-invasive imaging, and the angio-architecture of the target artery have the potential to assist in technique and device selection, which may improve FPE. The workshop participants support the study and validation of FPE as a benchmark to assess the effectiveness of EVT in stroke studies.

6. Imaging post reperfusion therapy

Brain imaging after EVT may help determine prognosis and adjuvant treatment. Perfusion imaging obtained after reperfusion therapy (medical therapy or EVT) can quantify the quality of macrovascular and microvascular reperfusion, blood-brain barrier (BBB) disruption, infarct evolution, and edema status. A study of CTP post-EVT determined the presence of hypoperfused brain tissue (Tmax > 6 sec) within 30 min of mechanical thrombectomy in most patients who achieved complete or angiographic reperfusion (mTICI 2a-3).⁵⁰ Even among patients who were deemed to have achieved complete angiographic reperfusion (mTICI 3), 42.5% demonstrated areas of cerebral hypoperfusion on post-thrombectomy perfusion imaging. Achieving recanalization after the first pass was associated with smaller volumes of hypoperfused tissue on post-EVT CTP, supporting the clinical benefit of first pass recanalization. A hypoperfusion volume <3.5 mL was independently associated with dramatic clinical recovery (OR, 4.1, 95% CI 2.0–8.3; P<0.01). Despite the profound effect of effective EVT on long-term functional outcome, a reduced infarct volume only accounts for approximately 12% of the treatment effect of EVT.^51,52 Thus, novel post-EVT imaging metrics may provide a better prediction of long-term outcome and identify opportunities for adjuvant therapy.

The status of the BBB can be assessed with brain MRI or NCCT. Disruption of the BBB due to ischemia can be seen as delayed gadolinium enhancement of CSF spaces (sulci) on fluid-attenuated inversion recovery (FLAIR) imaging.⁵³ This phenomenon has been named hyperintense acute reperfusion marker (HARM).⁵⁴ HARM has been associated with hemorrhagic transformation and worse clinical outcomes. On a post-procedural NCCT, disruption of the BBB can be seen as parenchymal hyperdensity.⁵⁵ The hyperdensity on NCCT likely represents extravasation of contrast medium into the extracellular spaces because of increased permeability of the BBB. This hyperdensity may be differentiated from the hyperdensity caused by hemorrhage, based on its selective localization in the gray matter (cortex or basal ganglia) and the absence or near-absence of mass effect on adjacent structures. Dual energy CT can be used to confirm this through iodine subtraction. A recent study identified BBB disruption in 58.2% (95% CI 51.4%-64.9%) of patients who underwent EVT.⁵⁶ Patients with BBB disruption had lower rates of early major neurologic improvement (8.6% vs 31.5%, p < 0.001), favorable outcome (39.8% vs 61.8%, p = 0.002), and higher rates of 90-day mortality (34.4% vs 14.6%, p = 0.001) and hemorrhagic complications (42.2% vs 8.7%, p < 0.001) than those without BBB disruption. Ng et al. assessed BBB disruption on a 24-hour post-procedure MRI. The study analyzed the associations between microvascular dysfunction in BBB disruption with ICH occurrence and edema formation in 238 patients.⁵⁷ Interestingly, BBB permeability was associated with worse outcomes and increased cerebral edema. The quantification of the degree and extent of BBB disruption and cerebral edema may be used to set blood pressure parameters, consider the intravenous infusion of hypertonic saline, or perform hemicraniectomy.

Another phenomenon that commonly can be observed on post-thrombectomy perfusion imaging is an increase in relative cerebral blood flow (rCBF) in tissue that was ischemic. Post-ischemic reactive hyperemia causes an approximately 57% increase in rCBF of the recanalized vascular territory that can last for a week after EVT.⁵⁸ This phenomenon may be related to the loss of cerebral autoregulation or hypermetabolism. An accurate quantification of hyperemia may be used in quantifying the response to cytoprotective agents. Disruption of the BBB coupled with hyperemia can lead to brain parenchyma edema, another phenomenon that can be observed on post-EVT imaging.⁵⁹

The workshop participants emphasized the need to quantify BBB disruption and characterize cerebral edema to optimize post-EVT management, and to measure the potential benefit of new cytoprotective treatments. Furthermore, there is a need for post-EVT imaging biomarkers of functional outcome.

OPTIMIZATION OF IMAGING TECHNOLOGY

7. Detection of early ischemia on NCCT

Detection of early ischemia on NCCT is notoriously difficult. Highly-trained readers have achieved a sensitivity of 43–71% in detecting early stroke (3–6 h) with NCCT, compared to 97% with diffusion-weighted imaging.⁶⁰ The ASPECTS was developed to simplify and standardize the rating of early ischemia on NCCT. The ASPECTS rating is based on a binary interpretation of ten regions within the MCA territory. For each region, the rater determines the presence or absence of hypoattenuation. At the extremes, patients with no hypoattenuation score a ten on the ASPECTS whereas patients with extensive early ischemia, involving all ten regions, score a zero. ASPECTS has been used extensively in clinical practice to triage patients with acute ischemic stroke for acute treatment and several trials have used ASPECTS in the selection of patients for EVT. For example, almost all endovascular trials that first demonstrated the efficacy of EVT, excluded patients with low (≤ 5) ASPECTS scores. In contrast, some recent trials that aimed to assess the effect of EVT in patients with large ischemic cores have recruited patients with low ASPECTS (3 to 5).^28,61 While ASPECTS has helped standardize the rating of early ischemic changes on NCCT, the interpretation of ASPECTS is variable, even between experts.^62,63 Key factors that contribute to the interrater variability in ASPECTS are the subtle nature of early ischemic changes on the NCCT, the lack of clearly defined boundaries of the ten ASPECTS regions and variation in the proportion of a region that is required to be abnormal in order to deduct a point.⁶⁴ Another limitation of ASPECTS is that the degree of hypoattenuation, a feature that might correlate with the reversibility of ischemic changes, is not captured in the score. Further, the ASPECTS regions differ in volume and therefore the ischemic core volume for a given ASPECT score can vary markedly.

The workshop participants identified the need for a reproducible volumetric method to describe the extent of early ischemic changes on NCCT. Commercially available software programs already exist for the automated qualitative evaluation of ASPECTS.^65,66 Bouslama et al reported that automated NCCT performs similarly to CTP in assessing post-reperfusion final infarct volume.⁶⁷ Recent data suggest that machine learning-NCCT estimated ischemic core is more accurate when obtained beyond one hour from stroke onset.⁶⁸ Furthermore, automated NCCT software can select patients with low likelihood of achieving a good outcome (e.g., ≥ 70 mL core at baseline) and who may not benefit from a transfer to a comprehensive stroke center for EVT.⁶⁹ While these programs reduce interrater variability, they do not address the need for a quantitative volumetric measure of early ischemia. New approaches can overcome this limitation. One method is the generation of relative NCCT maps using the hemisphere contralateral to the lesion as a reference. Hypodense brain tissue can be segmented based on its appearance, which can be measured in relative (e.g., more than 5% attenuation of the CT signal) or absolute (e.g., attenuation of more than 5 Hounsfield units of the CT signal) values. In addition to determining the location and volume of the early ischemic changes, the relative NCCT map can visualize the degree of hypoattenuation.⁷⁰ Another approach to quantify the degree of hypoattenuation on NCCT is the determination of net water uptake (NWU).^71,72 Broocks et al showed in a cohort of 254 patients that patients with low ASPECTS had elevated NWU and that the degree of net water uptake increased over time while ASPECTS did not change.⁷³

AI may help to overcome the limited accuracy of early stroke detection on NCCT. Early ischemic changes can be automatically detected using deep learning models.⁷⁴ AI-assisted stroke detection could also be advantageous in telemedicine approaches supporting non-primary stroke centers and could even be performed on mobile stroke units (MSU), thus potentially increasing access to reperfusion treatments and reducing time to treatment. AI is currently being extensively tested regarding its performance in early stroke detection. An AI model outperformed expert readers in detecting early ischemic changes on NCCT in a recent study,⁷⁵ and a systematic review including 11 studies and 1,976 cases revealed that AI-based ASPECTS performed similar or better than radiologists in identifying early stroke changes on NCCT.⁷⁶ Moreover, AI-based NCCT-ASPECTS was reported as good or better as human rating for posterior circulation stroke.⁷⁷ However, the accuracy and reliability of AI- and human based NCCT-ASPECTS depends on time from stroke onset to imaging and is lower in hyperacute stroke and fast stroke progressors.⁷⁸ Although AI-driven diagnostic processing is usually faster, it is not always superior to human rating, with AI showing less sensitivity in detecting LVO in CT angiography.⁷⁹ Implications of NCCT-ASPECTS using AI are unclear, AI may best be used under study protocols or under the supervision of human expert raters until proven clearly superior in a clinical setting. AI-assisted stroke diagnosis is feasible for all stroke imaging modalities, and the continuous evolvement of AI-based approaches is expected to result in significant performance improvements and wider applicability in stroke diagnosis.

8. Utility of cone beam computed tomography.

Cone beam CT (CBCT) imaging assessment of acute ischemic stroke patients with LVO in the angiography suite may improve stroke workflow and decrease time to recanalization. There are several advantages in obtaining a CBCT before EVT. Protocols that include the direct transfer to the angio-suite for EVT rely on CBCT to exclude hemorrhage and estimate the degree of early ischemic injury. The direct-to-angio approach may reduce time to treatment and functional outcomes.⁸⁰ In the ANGIOCAT (Direct to Angiography Suite without Stopping for Computed Tomography Imaging for patients with Acute Stroke) trial, CBCT was performed to exclude ICH or large established ischemic lesions that would contraindicate EVT.⁸¹ The study suggested better clinical outcomes in patients who were transferred directly to the angio-suite and who were imaged with CBCT compared to patients who underwent a conventional NCCT before going to the angiography suite. Improvements in clinical outcomes may have resulted from increased rates of successful EVT and shorter door-to-puncture times in the CBCT group. The higher rate of EVT in the CBCT group is likely the result of less stringent selection because CBCT imaging provides a less thorough parenchymal assessment than conventional NCCT imaging.

Compared to NCCT, CBCT imaging suffers from poorer delineation of the brain parenchyma and worse signal to noise which limits ischemia delineation.⁸² However, the new generation of CBCT exhibits better gray–white differentiation due to the high dynamic range flat detector, enabling four times more gray value differentiation, approaching the contrast resolution of conventional NCCT. Furthermore, the latest generation of X-ray tubes enables better penetration during the acquisition, especially in larger sized patients. In addition, it provides sharper images in all viewing directions. A study by Leyhe et al reported detection of ischemic lesions was feasible on CBCT scans with 71% sensitivity and 94% specificity (p<0.001; area under the curve 0.83, 95% CI 0.74 to 0.89) compared with NCCT scans. Additionally, ASPECTS ratings on CBCT showed a mean difference of only 0.5 points (95% CI 0.12 to 0.88) in the Bland–Altman plot compared with ratings of NCCT images.⁸³ Another study that compared the latest CBCT technology with NCCT, showed that early ischemic lesions were detected with a sensitivity of 73.3% and specificity of 94.7%, when compared to NCCT.⁸⁴ Further refinements in the X-ray tube trajectory to include caudal and cranial angulations have decreased artifacts.⁸⁵ Novel motion correction algorithms to improve imaging quality and diagnostic assessment of the brain parenchyma have been implemented.⁸⁶ Despite these advances, delineating grey-white matter differentiation and the visualization of infratentorial structures remains a limiting factor of CBCT. The workshop emphasized the need for collaboration between academia and industry to accelerate the development of high-definition CBCT. This advanced imaging technology holds significant promise and was recognized as a key area for focused efforts.

9. Imaging in mobile stroke units

A mobile stroke unit (MSU) was first implemented in 2008 in Germany, with the goal of prehospital care optimization.⁸⁷ MSUs are equipped with a CT scanner that can obtain a NCCT and a CTA. A MSU can triage patients and initiate thrombolytics. The first MSU trial conducted in Germany demonstrated shorter treatment times to intravenous thrombolysis (72 versus 153 min, p = 0.001). Approximately 57% of patients were treated with thrombolysis within one hour as compared to 4% of patients treated with standard management.⁸⁸ The benefits of MSU in screening, triaging, and treating patients with thrombolysis have been confirmed by several studies.^89,90 The BEST-MSU study confirmed the shorter administration of thrombolysis in patients screened and treated at a MSU versus conventional emergency medical services (72 vs 108 min respectively, p < 0.001).⁹¹ MSU management also resulted in significantly less disability at 90 days compared to conventional treatment (mean utility-weighted mRS score 0.72 vs. 0.66, p=0.002).

The workshop participants noted that MSUs are a positive addition to treating patients with AIS. However, technological and reimbursement challenges must be overcome to make this technology operational. In addition to the current uses of the MSU, the participants believed that “in the field” imaging of patients with acute stroke could facilitate cytoprotective studies. However, some technical developments are desired. Most MSUs utilize a portable 8-slice, CT scanner that can complete a NCCT of the brain and a CTA.⁹² Perfusion is limited to 1 cm slab which is unlikely to be clinically useful. CT scanners that allow for imaging of aortic arch and neck vessels are larger and therefore cannot be housed in a standard 12-foot ambulance. Workshop participants agree that further improvements in CT scanner miniaturization, while not compromising image resolution are needed. Similarly, optimizing the quality and capability of mobile MRI scanners is encouraged. Portable low-field MRI scanners are already in use at some hospitals, but the inability to acquire good-quality and rapid DWI scans is a limiting factor in the evaluation of stroke.⁹³ Finally, developing other technology for ischemic stroke identification, hemorrhagic stroke exclusion, and large-vessel occlusion detection in the field is similarly encouraged.

10. Vessel wall imaging to understand pathology.

Vascular pathology, such as cervical or intracranial atherosclerosis, is one of the most common causes of ischemic stroke worldwide. Historically, the imaging evaluation of atherosclerosis has focused on the assessment of the degree of luminal narrowing because more severe narrowing is associated with an increased risk of ischemic stroke.^94,95 However, 30–40% of patients who suffer an ischemic stroke do not have a clearly identifiable etiology for their stroke,⁹⁶and non-flow limiting atherosclerosis or other arterial vascular abnormalities may be the culprit in many of these patients.^97–100 The use of vessel wall imaging on 3-Tesla high-resolution magnetic resonance imaging is increasingly used to characterize stroke etiology.¹⁰¹ In one study, vessel wall imaging identified the probable cause of a patient’s stroke in 55% of cases, primarily by enhancing the detection of intracranial atherosclerotic disease.¹⁰² Similar studies have used high-resolution vessel wall imaging to identify the presence of underlying atherosclerosis and “culprit plaques” in patients previously deemed to have a cryptogenic stroke.¹⁰³ These data suggest that high resolution MRI and vessel wall imaging may add diagnostic value in ischemic stroke patients, but the adoption of these techniques has been relatively modest.

During the workshop, participants highlighted that lengthy acquisition times for vessel wall imaging have hindered its widespread adoption. They emphasized the importance of advancements in technology to shorten acquisition times. Additionally, the workshop underscored the necessity for high-quality prospective studies aimed at gaining a deeper understanding of the effectiveness of vessel wall imaging in characterizing intracranial atherosclerotic disease and other vasculopathies. By conducting such studies, we can enhance our knowledge and improve the utility of vessel wall imaging in clinical practice. Finally, clinical studies need to establish whether these imaging approaches have a utility in assessing patient response to statins and antiplatelet medications.

Conclusion

The participants in the neuroimaging workshop of STAIR XII have identified ten key areas of imaging that hold great promise in enhancing stroke outcomes. The development of novel imaging techniques and AI-based protocols aimed at early stroke detection through non-invasive imaging, identification of BBB damage following a stroke, characterization of clots before reperfusion, and vessel wall imaging for determining stroke etiology has garnered significant interest. Moreover, the implementation of newer CBCTs and MSUs could greatly enhance the workflow of LVO treatment. The workshop participants agreed on the importance of a collaborative effort involving investigators, industry, and regulators to advance imaging research and ultimately improve patient outcomes.

Non-standard Abbreviations and Acronyms:

NCCT

Non-contrast brain computed tomography

CTP

Computed tomography perfusion

LVO

Large vessel occlusion

FPE

First-pass effect

TICI

Thrombolysis in cerebral infarction

MT

Mechanical thrombectomy

mRS

Modified Rankin Score

ASPECTS

Alberta Stroke Program Early CT Score

BBB

Blood-brain barrier

MSU

Mobile stroke units

CBCT

Cone beam computed tomography

Appendix:

Other Acknowledged STAIR XII Participants

Didem Aksoy, Joe Broderick, Alicia C Castonguay, Supurna Ghosh, James C Grotta, George Harston, Gary R Houser, Kristopher Kuchenbecker, Lawrence L Latour, David S Liebeskind, John Kylan Lynch, Carolina Maier, Eva Mistry, J Mocco, Raul G Nogueira, Jeffrey L Saver, Marijke van Vlimmeren, Ajay K Wakhloo, and Lawrence Wechsler.