Large Vessel Occlusion Detection Research Articles

Evaluation of CINA® LVO artificial intelligence software for detection of large vessel occlusion in brain CT angiography

ObjectiveTo systematically evaluate the ability of the CINA® LVO software to detect large vessel occlusions eligible for mechanical thrombectomy on CTA using conventional neuroradiological assessment as gold standard. MethodsRetrospectively, two hundred consecutive patients referred for a brain CTA and two hundred patients that had been subject for endovascular thrombectomy, with an accessible preceding CTA, were assessed for large vessel occlusions (LVO) using the CINA® LVO software. The patients were sub-grouped by occlusion site. The original radiology report was used as ground truth and cases with disagreement were reassessed. Two-by-two tables were created and measures for LVO detection were calculated. ResultsA total of four-hundred patients were included; 221 LVOs were present in 215 patients (54%). The overall specificity was high for LVOs in the anterior circulation (93%). The overall sensitivity for LVOs in the anterior circulation was 54% with the highest sensitivity for the M1 segment of the middle cerebral artery (87%) and T-type internal carotid occlusions (84%). The sensitivity was low for occlusions in the M2 segment of the middle cerebral artery (13 and 0% for proximal and distal M2 occlusions respectively) and in posterior circulation occlusions (0%, not included in the intended use of the software). ConclusionsLVO detection sensitivity for the CINA® LVO software differs largely depending on the location of the occlusion, with low sensitivity for detection of some LVOs potentially eligible for mechanical thrombectomy. Further development of the software to increase sensitivity to all LVO locations would increase the clinical usefulness.

Noncontrast CT Selected Thrombectomy vs Medical Management for Late-Window Anterior Large Vessel Occlusion.

There is uncertainty whether patients with large vessel occlusion (LVO) presenting in the late 6-hour to 24-hour time window can be selected for endovascular therapy (EVT) by noncontrast CT (NCCT) and CT angiography (CTA) for LVO detection. We evaluated the clinical outcomes of patients selected for EVT by NCCT compared with those medically managed in the extended time window. This multinational cohort study was conducted at 66 sites across 10 countries. Consecutive patients with proximal anterior LVO stroke selected for EVT by NCCT or medically managed and presenting within 6-24 hours of time last seen well (TSLW) from January 2014 to May 2022 were included. The primary end point was the 90-day ordinal shift in the modified Rankin Scale (mRS) score. Inverse probability treatment weighting (IPTW) and multivariable methods were used. Of 5,098 patients screened, 839 patients were included, with a median (interquartile range) age of 75 (64-83) years; 455 (54.2%) were women. There were 616 patients selected to undergo EVT by NCCT (73.4%) and 223 (26.6%) who were medically managed. In IPTW analyses, there was a more favorable 90-day ordinal mRS shift in patients selected by NCCT to EVT vs those who were medically managed (odds ratio [OR] 1.99, 95% CI 1.53-2.59; p < 0.001). There were higher rates of 90-day functional independence (mRS 0-2) in the EVT group (40.1% vs 18.4%, OR 3.31, 95% CI 2.11-5.20; p < 0.001). sICH was nonsignificantly higher in the EVT group (8.5% vs 1.4%, OR 3.77, 95% CI 0.72-19.7, p = 0.12). Mortality at 90 days was lower in the EVT vs MM group (23.9% vs 32.3%, OR 0.61, 95% CI 0.45-0.83, p = 0.002). In patients with proximal anterior LVO in the extended time window, there was a lower rate of disability and mortality in patients selected with NCCT and CTA to EVT compared with those who were medically managed. These findings support the use of NCCT as a simpler and more inclusive approach to patient selection in the extended window. This study was registered at ClinicalTrials.gov under NCT04096248. This study provides Class III evidence that for patients with proximal anterior circulation occlusion presenting with ischemic stroke from 6 to 24 hours, compared with medical management, those undergoing thrombectomy based on NCCT have reduced disability and mortality at 90 days.

Prospective Validation of Glial Fibrillary Acidic Protein, d ‐Dimer, and Clinical Scales for Acute Large‐Vessel Occlusion Ischemic Stroke Detection

BACKGROUND Large‐vessel occlusion (LVO) ischemic stroke is responsible for significant morbidity and mortality. We have previously described a novel tool for acute LVO detection that combines blood‐based biomarkers (glial fibrillary acidic protein and d ‐dimer) with stroke severity scales to achieve high accuracy. Accordingly, the present study sought to prospectively validate cutoff values that we had previously established for biomarkers and scales. METHODS The TIME (Testing for Identification Markers of Stroke) trial was designed as a prospective observational diagnostic accuracy study. All ambulance‐identified stroke code activations &lt;18 hours from symptom onset were recruited at Brandon Regional Hospital (Brandon, FL) between May 2021 and August 2022. Previously determined cutoff concentrations of plasma glial fibrillary acidic protein (213 pg/mL) and d ‐dimer (600 ng/mL) were used in combination with prehospital stroke scales to detect LVO. We compared rates of LVO detection against a reference standard using computed tomography/magnetic resonance angiography. RESULTS A total of 382 patients with suspected stroke were recruited. The final cohort was composed of 323 patients with suspected stroke with the following distribution: LVO ischemic stroke (n = 29, 9%), non‐LVO ischemic stroke (n = 48, 15%), hemorrhagic stroke (n = 13, 4%), transient ischemic attack (n = 12, 3.9%), and stroke mimics (n = 220, 68.1%). Combining blood‐based biomarkers (glial fibrillary acidic protein and d ‐dimer) with the scale field assessment stroke triage for emergency destination yielded the best performance for LVO detection, with specificity of 94% and sensitivity of 71%. Performance was found to be higher in a subanalysis focusing on patients presenting &lt;6 hours from symptom onset, with 93% specificity and 81% sensitivity. Critically, application of the biomarker and stroke scale algorithms ruled out all patients with hemorrhage. CONCLUSION The present work prospectively validated the potential utility of previously defined glial fibrillary acidic protein and d ‐dimer cutoff levels (ie, 213 pg/mL and 600 ng/mL, respectively), demonstrating their value for discrimination of LVO stroke from differential diagnoses during code stroke workups. (ClinicalTrials.gov number, NCT04292600.)

Headpulse measurement can reliably identify large-vessel occlusion stroke in prehospital suspected stroke patients: Results from the EPISODE-PS-COVID study.

Large-vessel occlusion (LVO) stroke represents one-third of acute ischemic stroke (AIS) in the United States but causes two-thirds of poststroke dependence and >90% of poststroke mortality. Prehospital LVO stroke detection permits efficient emergency medical systems (EMS) transport to an endovascular thrombectomy (EVT)-capable center. Our primary objective was to determine the feasibility of using a cranial accelerometry (CA) headset device for prehospital LVO stroke detection. Our secondary objective was development of an algorithm capable of distinguishing LVO stroke from other conditions. We prospectively enrolled consecutive adult patients suspected of acute stroke from 11 study hospitals in four different U.S. geographical regions over a 21-month period. Patients received device placement by prehospital EMS personnel. Headset data were matched with clinical data following informed consent. LVO stroke diagnosis was determined by medical chart review. The device was trained using device data and Los Angeles Motor Scale (LAMS) examination components. A binary threshold was selected for comparison of device performance to LAMS scores. A total of 594 subjects were enrolled, including 183 subjects who received the second-generation device. Usable data were captured in 158 patients (86.3%). Study subjects were 53% female and 56% Black/African American, with median age 69 years. Twenty-six (16.4%) patients had LVO and 132 (83.6%) were not LVO (not-LVO AIS, 33; intracerebral hemorrhage, nine; stroke mimics, 90). COVID-19 testing and positivity rates (10.6%) were not different between groups. We found a sensitivity of 38.5% and specificity of 82.7% for LAMS ≥ 4 in detecting LVO stroke versus a sensitivity of 84.6% (p < 0.0015 for superiority) and specificity of 82.6% (p = 0.81 for superiority) for the device algorithm (CA + LAMS). Obtaining adequate recordings with a CA headset is highly feasible in the prehospital environment. Use of the device algorithm incorporating both CA and LAMS data for LVO detection resulted in significantly higher sensitivity without reduced specificity when compared to the use of LAMS alone.

Large vessel occlusion detection by non-contrast CT using artificial ıntelligence.

Computer vision models have been used to diagnose some disorders using computer tomography (CT) and magnetic resonance (MR) images. In this work, our objective is to detect large and small brain vessel occlusion using a deep feature engineering model in acute of ischemic stroke. We use our dataset. which contains 324 patient's CT images with two classes; these classes are large and small brain vessel occlusion. We divided the collected image into horizontal and vertical patches. Then, pretrained AlexNet was utilized to extract deep features. Here, fc6 and fc7 (sixth and seventh fully connected layers) layers have been used to extract deep features from the created patches. The generated features from patches have been concatenated/merged to generate the final feature vector. In order to select the best combination from the generated final feature vector, an iterative selector (iterative neighborhood component analysis-INCA) has been used, and this selector has chosen 43 features. These 43 features have been used for classification. In the last phase, we used a kNN classifier with tenfold cross-validation. By using 43 features and a kNN classifier, our AlexNet-based deep feature engineering model surprisingly attained 100% classification accuracy. The obtained perfect classification performance clearly demonstrated that our proposal could separate large and small brain vessel occlusion detection in non-contrast CT images. In this aspect, this model can assist neurology experts with the early recanalization chance.

Deep learning-assisted diagnosis of large vessel occlusion in acute ischemic stroke based on four-dimensional computed tomography angiography.

To develop deep learning models based on four-dimensional computed tomography angiography (4D-CTA) images for automatic detection of large vessel occlusion (LVO) in the anterior circulation that cause acute ischemic stroke. This retrospective study included 104 LVO patients and 105 non-LVO patients for deep learning models development. Another 30 LVO patients and 31 non-LVO patients formed the time-independent validation set. Four phases of 4D-CTA (arterial phase P1, arterial-venous phase P2, venous phase P3 and late venous phase P4) were arranged and combined and two input methods was used: combined input and superimposed input. Totally 26 models were constructed using a modified HRNet network. Assessment metrics included the areas under the curve (AUC), accuracy, sensitivity, specificity and F1 score. Kappa analysis was performed to assess inter-rater agreement between the best model and radiologists of different seniority. The P1 + P2 model (combined input) had the best diagnostic performance. In the internal validation set, the AUC was 0.975 (95%CI: 0.878-0.999), accuracy was 0.911, sensitivity was 0.889, specificity was 0.944, and the F1 score was 0.909. In the time-independent validation set, the model demonstrated consistently high performance with an AUC of 0.942 (95%CI: 0.851-0.986), accuracy of 0.902, sensitivity of 0.867, specificity of 0.935, and an F1 score of 0.901. The best model showed strong consistency with the diagnostic efficacy of three radiologists of different seniority (k = 0.84, 0.80, 0.70, respectively). The deep learning model, using combined arterial and arterial-venous phase, was highly effective in detecting LVO, alerting radiologists to speed up the diagnosis.

Machine Learning–Enabled Automated Large Vessel Occlusion Detection Improves Transfer Times at Primary Stroke Centers

Background Accelerating door‐in‐door‐out (DIDO) times at primary stroke centers (PSCs) for patients with large vessel occlusion (LVO) acute ischemic stroke transferred for possible endovascular stroke therapy (EVT) is important to optimize outcomes. Here, we assess whether automated LVO detection coupled with secure communication at non‐EVT performing PSCs improves DIDO time and increases the proportion of patients receiving EVT after transfer. Methods From our prospectively collected multicenter registry, we identified patients with LVO acute ischemic stroke that presented to one of 7 PSCs in the Greater Houston area from January 1, 2021, to February 27, 2022. Noncontrast computed tomography and computed tomographic angiography were performed in all patients at the time of presentation, per standard of care. A machine learning (artificial intelligence [AI]) algorithm trained to detect LVO (Viz.AI) from computed tomographic angiography was implemented at all 7 hospitals. The primary outcome of the study was DIDO at the PSCs and was determined using multivariable linear regression adjusted for sex and on/off hours. Secondary outcomes included likelihood of receiving EVT post‐transfer. Results Among 115 patients who met inclusion criteria, 80 were evaluated pre‐AI and 35 post‐AI. The most common occlusion locations were middle cerebral artery (51.3%) and internal carotid artery (25.2%). There were no substantial differences in demographics or presentation characteristics between the 2 groups. Median time from onset to PSC arrival was 117 minutes (interquartile range, 54–521 minutes). In univariable analysis, patients evaluated at the PSCs after AI implementation had a shorter DIDO time (median difference, 77 minutes; P &lt;0.001). In multivariable linear regression, patients evaluated with automated LVO detection AI software were associated with a 106‐minute (95% CI, −165 to −48 minutes) reduction in DIDO time but no difference in likelihood of EVT post‐transfer (odd ratio, 2.13 [95% CI, 0.88–5.13). Conclusion Implementation of a machine learning method for automated LVO detection coupled with secure communication resulted in a substantial decrease in DIDO time at non‐EVT performing PSCs.

Portable cerebral blood flow monitor to detect large vessel occlusion in patients with suspected stroke

BackgroundEarly detection of large vessel occlusion (LVO) facilitates triage to an appropriate stroke center to reduce treatment times and improve outcomes. Prehospital stroke scales are not sufficiently sensitive, so we...

Door to Puncture in Large Vessel Occlusions Pre‐ and Postimplementation of an Automated Image Interpretation and Communication Platform: A Single Center Study

Background Artificial intelligence platforms, like Viz.ai with large vessel occlusion detection, have been used for disease detection and interprovider communication. Whether this software expedites patient transfer and evaluation for treatment needs further exploration. Methods A single‐center retrospective registry was queried for patients with acute large vessel occlusion of the intracranial internal carotid, middle cerebral M1 or M2 segments, or basilar artery treated in a comprehensive stroke network (8 spokes, 1 hub) for 6 months pre‐ and post‐implementation of the Viz large vessel occlusion platform (excluding a 1‐month “washout” period). Robust regression was used to summarize time from initial hospital contact to arterial puncture (primary outcome) between periods, with prespecified subgroup analyses, which were assessed using interaction terms. Results Of the 132 patients (n = 58 preintervention), there were nonsignificantly fewer patients undergoing endovascular therapy in the postintervention period (86.2% preintervention versus 73.0% postintervention; P = 0.07). Among patients who underwent endovascular therapy (n = 50 preintervention, n = 54 postintervention), there was a nonsignificant reduction in time from first contact to arterial puncture (median 155 minute preintervention versus 116 minute postintervention; P = 0.10); however, this became significant in adjusted robust regression accounting for stroke severity, age, Alberta Stroke Program Early Computed Tomography Scale score, daytime versus nighttime and weekend versus weekday arrival, and use of perfusion imaging (β −20.9 [95% CI, −40.5 to −1.4)]. There was also a significant interaction observed for the association between spoke versus hub arrival and the Viz large vessel occlusion period, with shorter intervals observed for transferred patients (n = 32 preintervention with a median of 169 versus 142 minutes for n = 33 postintervention; P interaction &lt;0.01). Conclusion Implementation of the artificial intelligence platform was not associated with shorter intervals between initial hospital contact and neurointervention among all‐comers. A meaningful difference in time to treatment was observed among transferred patients. Larger data sets are needed to validate these observations.

Abstract 44: Portable Optical Blood Flow Monitor for the Detection of Large Vessel Occlusion

Introduction: Early detection of large vessel occlusion (LVO) facilitates triage to appropriate stroke centers and reduces treatment times. Prehospital stroke scales are insensitive, so we investigated the ability of the Openwater Headset, a portable optical blood flow monitor, to detect LVO in patients undergoing acute stroke evaluation. Hypothesis: We hypothesized that the Openwater Headset would detect LVO with greater sensitivity and specificity than prehospital scales. Methods: At two comprehensive stroke centers, eligible patients presented to the emergency department or were transferred from another facility for acute stroke evaluation within 24 hours of symptom onset and NIHSS ≥2. Eligible patients underwent emergent vascular imaging as per routine care. LVO was defined as occlusion of the ICA or the first/second segment of the MCA. Each patient underwent a 70-second bedside optical blood flow evaluation with the Openwater Headset (prior to endovascular therapy, if applicable). The Openwater Headset generated blood flow waveforms based on the relative change in the speckle contrast, acquired at 40 Hz. A deep neural network was trained to identify waveform features that discriminate LVO status ( Figure 1a ), withholding a validation set to test model performance. ROC analysis compared Openwater Headset performance with that of prehospital stroke scales. Results: 135 patients completed the optical evaluation a median of 6 hours after symptom onset (IQR: 3-14). The median NIHSS was 9 (IQR: 4-18). 52 patients (39%) had a LVO. The Openwater LVO stroke alert algorithm had 81% sensitivity and 80% specificity for the presence of LVO (AUC 0.85 by ROC analysis), exceeding that of prehospital stroke scales ( Figure 1b ). Conclusions: The Openwater Headset outperformed prehospital stoke scales for LVO detection in patients who underwent acute stroke evaluation, but further investigation is needed to assess performance in pre-hospital settings.

Abstract TMP57: User Engagement is Associated With Clinical Impacts of Automated LVO Detection Software

Introduction: Machine learning (ML)-based acute ischemic stroke (AIS) large vessel occlusion (LVO) detection algorithms can reduce door to groin time (DTG). Here, we examine whether the dependency of these benefits on end user engagement and interaction. Methods: This analysis was conducted as a pre-planned post hoc analysis of our multicenter, prospective randomized clinical trial (NCT05838456). ML-based LVO detection software was implemented at 4 comprehensive stroke centers in the greater Houston area in a stepped fashion between Jan 2021 and March 2022. Patients were included in this analysis if they underwent EVT for LVO AIS. Exclusion criteria included inter-hospital transfers or inpatient stroke alerts. ML-software utilization was quantified as the number of software interactions including imaging viewing and/or HIPAA-compliant text messaging and was trichotomized at the hospital level into low/medium/high. Primary outcome was the reduction in DTG relation to pre-ML implementation by hospital utilization level. Secondary outcomes included CT to groin puncture time, and sensitivity analyses in subsets of patients who did and did not have an LVO alert sent. Results: Among 243 patients that met inclusion criteria, median age was 70 (IQR 58-79), 50% were female and median NIHSS was 17 (IQR 11-22). There were 2 high-utilization centers interacting with the ML software mean 160 times, one medium center that interacted 64 times and one low center interacting 42 times during the study period. We observed a reduction on DTG of 11 minutes in the high-interaction center (p&lt;0.01) but no significant reduction in DTG in the medium or low-interaction centers. Similarly, time from CT scan initiation to groin puncture fell in the high-utilization centers by 32 minutes (p&lt;0.05) but no significant change in others. Without adjusting for hospital-level software utilization, there was no statistically significant reduction in DTG for patients analyzed by the ML software versus those that were not (p=0.35). Conclusion: The benefit of ML-enhanced LVO AIS workflows are dependent on user engagement.

Abstract WP70: Novel Weighted FANG-D Scale Scoring to Detect Large Vessel Occlusion

Objective: To determine if a novel weighted FANG-D scoring scale (NWFSS) increases sensitivity and specificity rates of large vessel occlusion (LVO) as compared to those reported with the original FANG-D scale scoring. Methodology: The present study used an observational, descriptive design to identify the important variables (field cut, aphasia, neglect, gaze preference, and dense hemiparesis) associated with the likelihood of detecting LVO. We used the FANG-D scale in patients admitted through the emergency department (ED) of a comprehensive stroke center (CSC), from December 2022 through April 2023. The NIHSS was performed and compared with the NWFSS for analysis. Each component was required to have a score ≥ 2(moderate to severe deficits) except dense hemiparesis which required a score ≥3 (significant weakness) to receive one point on the FANG-D scale. A total score of 2 or higher on FANG-D indicated a positive result suggesting LVO. If aphasia or neglect were present with a NIHSS of≥2, it was considered a positive FANG-D score. Results: A total of 311 patients were included in the study for analysis with a mean age of 67 ±16.4 years of whom 56.6% were female. The sensitivity and specificity rates were 85.7% (Confidence Interval (CI) =72.8%-94.1%) and 90.5% (CI= 86.2%-93.7%) respectively for positive predictive value of NWFSS to identify LVO. The sensitivity and specificity of the NIHSS has been reported as 86% and 60% respectively for LVO detection. The previous FANG-D study group used lower values in calculating the FANG-D scores and reported a sensitivity rate of 98% and specificity of 76.7%. Conclusion: The NWFSS has high specificity and is reliable for LVO prediction. While the NIHSS assesses eleven items sensitive for stroke diagnosis, it may take up to 10 minutes to complete and is not specific for LVO, whereas the FANG-D assesses 5items critical to LVO detection and is simple and fast with completion in &lt;1 minute, minimizing time to intervention. Our study occurred in the ED, but this scale scoring method could be tested in the prehospital setting. The use of a scale with higher specificity for LVO could improve patient outcomes and resource utilization by faster routing of patients to CSCs capable of performing mechanical thrombectomy.

Abstract 90: Optimizing Triage Assessments for Large and Medium Vessel Occlusions

Introduction: Stroke symptom assessment tools are used in various settings, including telestroke triage, to identify patients who may be candidates for endovascular therapy (EVT). EVT is now being offered to more patients including those with medium vessel occlusions (MeVOs). We hypothesize that triage screening tools have differing rates of detection for large vessel occlusions (LVOs) versus medium vessel occlusions (MeVOs). Methods: Using our 2021 stroke registry, we created a retrospective cohort of ischemic stroke patients with LVO, MeVO or no occlusion. Patients were treated with the standard of care (e.g., thrombolytics, EVT). Clinical and radiographic data were abstracted for each case. Patients were evaluated for hemiparesis (HP) and vision, aphasia, or neglect (VAN) using the documented presenting exam. ROC analysis was conducted to evaluate the ability of the HP + VAN screening tool to detect LVOs and MeVOs. Results: The cohort consisted of 299 patients (median age 66 years, 48% female, 36% Black, median NIHSS score 9). Among the 175 patients with recorded HP + one of the VAN signs, 81% (n=141) had an occlusion (LVO or MeVO). HP+VAN criteria was not met by 11% of LVOs and 33% of MeVOs. HP + VAN tool performed best for the combined endpoint of LVO or MeVO (sensitivity 0.83, specificity 0.74, AUC 0.786, Figure). Variations of this tool did not significantly improve performance (additional models with AUCs will be presented) with the exception of changing the score to from HP+VAN to “HP or VAN” (sensitivity 1.00, specificity 0.25). Conclusions: In our sample from a tertiary care center, HP+VAN is a reasonable tool for the detection of LVOs or MeVOs. The only variation that improved sensitivity was screening for hemiparesis or VAN. While this prevented us from missing any LVOs or MeVOs, this approach comes at a heavy cost of reduced specificity.

Abstract WP92: Application of a Horizontal Communication Process in Combination With Artificial Intelligence (AI) to Improve Stroke Care in Patients With Large Vessel Occlusion (LVO)

Introduction: Timely acute stroke care in patients with large vessel occlusion (LVO) requires both early detection of LVO and efficient multidisciplinary communication. In this study, we aimed to evaluate the impact of using AI and a horizontal communication system within the RapidAI application platform on acute stroke quality metrics in patients with LVO. Methods: On April 2022, the RapidAI application was launched at our center, which allowed for use of AI as a screening measure for LVO detection with immediate notification to neurology, neuroradiology, and neuro-interventional teams. The various teams used instant messaging within the application to communicate clinical data, imaging findings and shared clinical decision making. In this single center study, we retrospectively evaluated stroke care quality metrics for 167 consecutive patients (mean age: 71.6, 49.1% male) who underwent mechanical thrombectomy (MT) from January 2021 to May 2023. Data were compiled from multiple resources including Get-With-The-Guidelines®, chart abstraction, and internal healthcare databases. The groups were compared using appropriate statistical tests and significance was determined using a two-sided alpha of 0.05. Results: 85 patients underwent MT before implementation of the new system, and 82 patients underwent MT after. Demographic features (age, sex, race), cerebrovascular risk factors (hypertension, diabetes, hyperlipidemia, smoking, BMI) and stroke severity were similar between the two groups (p&gt;0.05). Average door to groin puncture (pre: 97 minutes, post: 82 minutes, p=0.10) and discharge disposition (p=0.02) were significantly different between the two groups. In an adjusted analysis of door to groin puncture time, the difference between pre-RapidAI and post-RapidAI was an average reduction of 19 minutes trending towards significance (p = 0.06). Conclusion: Application of a horizontal communication platform in combination with RapidAI was associated with significant improvements in time to intervention in patients with LVO and a significant improvement in the number of patients discharged home.

Abstract WP67: Utilization and Impact of Large Vessel Occlusion Stroke Screening Practices and Selected Target Stroke III Strategies in the Get-with-the-Guidelines Registry Hospitals

Background: The utilization and impact of pre-hospital and in-hospital large vessel occlusion (LVO) stroke screening protocols and Target: Stroke phase III (TS III) best-practice strategies on time metrics for endovascular treatment at a national level, have not been studied. Methods: We sent an online survey in July 2022 to hospital representatives of 2528 hospitals participating in the GWTG registry about their EMS systems, clinical and imaging protocols, and utilization of selected TS III best-practice strategies (multiple choice, 0-100 scale, and yes/no questions). We obtained Individual patient-level data from the GWTG-Stroke Registry from January 2017 to March 2022. Multivariable linear regression models were performed to investigate the associations of these strategies with door-to-puncture (DTP) in endovascular (EVT) patients. Results: Out of 2455 sites that met our inclusion criteria, 1455 sites completed the survey, with a response rate of 59.3%. Hospital-level baseline characteristics, utilization of selected LVO screening practices, and site-reported strategies associated with shorter DTP times are shown (Table 1). Strategies associated with shorter DTP were the performance of simultaneous vascular imaging along with non-contrast CT scan on all stroke patients within 24 hours, and the use of newer technologies for LVO detection in the field leading to a 7.1 min (CI -12.8, -1.5) decrease and a 10 min (CI -18.6, -1.3) decrease in DTP with every 25% increase in the utilization of these strategies, respectively (Table 1). Conclusion: The simultaneous performance of vascular imaging with non-contrast CT scan in all stroke patients presenting within 24 hours and the use of newer technologies for LVO detection in the field could lead to shorter DTP times.

Abstract WMP43: Disparities in Large Vessel Occlusion Stroke Imaging and Endovascular Treatment Metrics, and Procedural Exclusions in the Global Registry of Stroke Care Quality (Res-Q)

Introduction: Global access to mechanical thrombectomy for large vessel occlusion (LVO) stroke is low and disparate between regions. Comparative global data on time metrics for imaging and endovascular treatment (EVT) performance, as well as the reasons for excluding EVT, are lacking. Methods: This is a cross-sectional study using quality metrics from the Res-Q registry on patients presenting directly to the hospital with acute ischemic stroke from January 1, 2022, to December 31, 2022, in countries with 200 or more cases. We used descriptive statistical methods to study the time metrics of interest including door to imaging and door to puncture. Additionally, reasons for not performing thrombectomy in patients with LVO were collected (Table 1). Results: Among the 153,181 patients from 912 hospitals across 67 countries enrolled in the Res-Q registry, 125,390 had an acute ischemic stroke. After excluding secondary transfers (n=26,648), patients with missing or erroneous data (n=42,232), and countries with less than 200 cases (n=1,626), data from a total of 54,884 patients from 631 hospitals across 30 countries were analyzed. CT angiography (CTA) was performed in 24,215 (44.1%) patients and EVT was done in 3,649 (6.6%) patients. For patients who had simultaneous CT+CTA, the median door-to-CTA time was 27.5 min (IQR 22.5-33.9) and the median door-to-puncture time was 98.5 (IQR 79.8-129.3) minutes. The most prevalent specific reason for excluding EVT in patients with LVO on CTA/MRA was the presentation in the late time window (42.0%). Only 11 (36.7%) countries achieved a median door-to-puncture time within 90 minutes (Table 1). Conclusions: The performance of acute imaging for LVO detection and time metrics for EVT appear to be disparate between countries and warrant further study. Additional data from global stroke care quality registries are needed to set benchmarks, compare regional stroke systems of care, and identify gaps to mobilize resources appropriately.

Rapid Arterial Occlusion Evaluation (RACE) Tool in Detecting Large Cerebral Vessel Occlusions; a Systematic Review and Meta-Analysis.

Large vessel occlusion (LVO) strokes are linked to higher mortality rates and a greater risk of long-term disability. This study aimed to evaluate the diagnostic performance of the Rapid Arterial Occlusion Evaluation (RACE) tool in detecting LVO through a systematic review and meta-analysis. A comprehensive search was conducted across online databases including PubMed, Embase, Scopus, and Web of Science, up to June 25th, 2023. Additionally, a manual search on Google and Google Scholar was performed to identify studies that assessed the diagnostic accuracy of the RACE scale in detecting LVO among patients with stroke symptoms. Data extracted from 43 studies were analyzed. The optimal cut-off points were determined to be 3 and 4, with a sensitivity of 0.86 (95% confidence interval (CI): 0.78, 0.91) and specificity of 0.57 (95% CI: 0.49, 0.67) for cut-off ≥3, and a sensitivity of 0.78 (95% CI: 0.70, 0.84) and specificity of 0.68 (95% CI: 0.59, 0.75) for cut-off ≥4. Subgroup meta-regression analysis revealed significant variations in sensitivity and specificity. RACE scale's sensitivity was significantly higher in LVO detection in suspected stroke cases, in pre-hospital settings, prospective design studies, and when considering both anterior and posterior occlusions for LVO definition. RACE scale's specificity was significantly higher when evaluating confirmed stroke cases, in-hospital settings, and considering only anterior occlusions for LVO definition and retrospective design studies. Notably, RACE exhibited higher sensitivity and specificity when utilized by neurologists and physicians compared to other emergency staff. Despite these variations, our study found comparable diagnostic accuracy across different conditions. A high level of evidence indicates that the RACE scale lacks promising diagnostic value for detection of LVOs. A sensitivity range of 0.69 to 0.86 is insufficient for a screening tool intended to aid in the diagnosis of strokes, considering the substantial morbidity and mortality associated with this condition.

Deep Learning-Based Synthetic TOF-MRA Generation Using Time-Resolved MRA in Fast Stroke Imaging.

Time-resolved MRA enables collateral evaluation in acute ischemic stroke with large-vessel occlusion; however, a low SNR and spatial resolution impede the diagnosis of vascular occlusion. We developed a CycleGAN-based deep learning model to generate high-resolution synthetic TOF-MRA images using time-resolved MRA and evaluated its image quality and clinical efficacy. This retrospective, single-center study included 397 patients who underwent both TOF- and time-resolved MRA between April 2021 and January 2022. Patients were divided into 2 groups for model development and image-quality validation. Image quality was evaluated qualitatively and quantitatively with 3 sequences. A multireader diagnostic optimality evaluation was performed by 16 radiologists. For clinical validation, we evaluated 123 patients who underwent fast stroke MR imaging to assess acute ischemic stroke. The diagnostic confidence level and decision time for large-vessel occlusion were also evaluated. Median values of overall image quality, noise, sharpness, venous contamination, and SNR for M1, M2, the basilar artery, and posterior cerebral artery are better with synthetic TOF than with time-resolved MRA. However, with respect to real TOF, synthetic TOF presents worse median values of overall image quality, sharpness, vascular conspicuity, and SNR for M3, the basilar artery, and the posterior cerebral artery. During the multireader evaluation, radiologists could not discriminate synthetic TOF images from TOF images. During clinical validation, both readers demonstrated increases in diagnostic confidence levels and decreases in decision time. A CycleGAN-based deep learning model was developed to generate synthetic TOF from time-resolved MRA. Synthetic TOF can potentially assist in the detection of large-vessel occlusion in stroke centers using time-resolved MRA.

Abstract 153: Pre‐procedural 3D visualization of deep‐learning vascular segmentation enhances identification of difficult femoral access in thrombectomy

Introduction Difficult or impossible femoral access remains a significant burden for mechanical thrombectomy (MT) in stroke, affecting between 4.4% (impossible) to ∼10% of cases (time to first angiography [T1A] &gt; 30 min) [1]. Visualization of CT angiography (CTA) or MR angiography are the only available diagnostic tools for neurointerventionalists (NIs) to qualitatively assess catheterization difficulty ahead of intervention. Visualization of 3D segmentation of the head‐and‐neck arteries from CTA may provide enhanced spatial information for catheter access difficulty assessment to the occlusion site, as well as large vessel occlusion (LVO) localization or carotid stenosis detection. Methods Three observers (2 expert NIs and 1 medical image engineer) assessed a sample of patients with anterior circulation stroke rating LVO location, femoral access difficulty according to a Likert scale (0 to 5) and assessment of radial access compared to femoral (easier or not). Raters were asked to answer the same form for each case twice, once after visualizing only the CTA and again after visualizing a deep‐learning, automated 3D segmentation of the head‐and‐neck arteries from CTA [2]. Cases were randomly sampled and blind to raters for both assessments. Likert scale values were normalized and averaged across observers, and correlation to T1A was studied. LVO localization accuracy and time needed for the analysis was also assessed. Results The final sample included N = 117 cases, where 22.98% presented difficult or impossible access. LVO location prevalence was 13/15/45/27% for extracranial ICA (eICA)/TICA/M1/M2, and 49/51% for right/left. Averaged Likert values presented significant linear correlation for both segmentation and CTA visualization, but this was much stronger in the former (R: 0.58 vs 0.30, p&lt;0.001). Three‐dimensional visualization of segmented vessels allowed for an enhanced detection of difficult or impossible femoral access (ROC‐AUC: 0.85 vs 0.66). LVO detection was also superior after segmentation visualization (mean accuracy across observers: 0.81 vs 0.72) and time needed for assessment was on average 2.3 times shorter (44 s vs 100 s). In difficult or impossible patients, radial access was considered easier in 74% (63‐89%) of cases when visualizing the segmentation compared to 42% (35‐48%) when inspecting the CTA. Conclusion Visualization of deep‐learning 3D arterial segmentation resulted in enhanced difficult or impossible femoral access for MT, LVO detection and faster analysis over traditional pre‐procedural CTA inspection. Adding this visualization to the current workflow could enhance difficult access detection ahead of intervention.

Abstract 035: Workflow Improvements from Automated LVO Detection Algorithms Are Dependent on Care Team Engagement

Introduction Automated machine learning (ML)‐based large vessel occlusion (LVO) detection algorithms have been shown to improve in‐hospital workflow metrics including door to groin time (DTG). The degree to which care team engagement and interaction are required for these benefits remains incompletely characterized. Methods This analysis was conducted as a pre‐planned post hoc analysis of our multicenter, prospective randomized clinical trial (NCT05838456). ML‐based LVO detection software was implemented at 4 comprehensive stroke centers in the greater Houston area in a stepped fashion between Jan 2021 and March 2022. Patients were included in this analysis if they underwent EVT for LVO AIS. Patients were excluded if they presented as inter‐hospital transfers or inpatient stroke alerts, as these workflows diverge considerably from ER‐based ones. ML‐software utilization quantified as the number of software interactions including imaging viewing and/or HIPAA‐compliant text messaging and was trichotomized at the hospital level into low/medium/high. Primary outcome was the reduction in DTG relation to pre‐ML implementation by hospital utilization level. Secondary outcomes included CT to groin puncture time, and sensitivity analyses in subsets of patients who did and did not have an LVO alert sent. Results Among 243 patients that met inclusion criteria, median age was 70 (IQR 58‐79), 50% were female and median NIHSS was 17 (IQR 11‐22). ML‐software utilization varied considerably with high‐utilization centers (n=2) interacting with mean 160 times, medium (n=1) 64 times and low center (n=1) 42 times during the study period. We observed a reduction on DTG of 11 minutes in the high‐interaction center (p&lt;0.01) but no significant reduction in DTG in the medium or low‐interaction centers. Similarly, time from CT scan initiation to groin puncture fell in the high‐utilization centers by 32 minutes (p&lt;0.05) but no significant change in others. Without adjusting for hospital‐level software utilization, there was no statistically significant reduction in DTG for patients analyzed by the ML software versus those that were not (p=0.35). Conclusion ML‐based workflow improvements are dependent on care team adoption and utilization.