Identification Of Large Vessel Occlusion Research Articles

Machine learning models improve prediction of large vessel occlusion and mechanical thrombectomy candidacy in acute ischemic stroke

Background and PurposeEarly identification of large vessel occlusions (LVO) and timely recanalization are paramount to improved clinical outcomes in acute ischemic stroke. A stroke assessment that maximizes sensitivity and specificity for LVOs is needed to identify these cases and not overburden the health system with unnecessary transfers. Machine learning techniques are being used for predictive modeling in many aspects of stroke care and may have potential in predicting LVO presence and mechanical thrombectomy (MT) candidacy. MethodsIschemic stroke patients treated at Loyola University Medical Center from July 2018 to June 2019 (N = 286) were included. Thirty-five clinical and demographic variables were analyzed using machine learning algorithms, including logistic regression, extreme gradient boosting, random forest (RF), and decision trees to build models predictive of LVO presence and MT candidacy by area of the curve (AUC) analysis. The best performing model was compared with prior stroke scales. ResultsWhen using all 35 variables, RF best predicted LVO presence (AUC = 0.907 ± 0.856–0.957) while logistic regression best predicted MT candidacy (AUC = 0.930 ± 0.886–0.974). When compact models were evaluated, a 10-feature RF model best predicted LVO (AUC = 0.841 ± 0.778–0.904) and an 8-feature RF model best predicted MT candidacy (AUC = 0.862 ± 0.782–0.942). The compact RF models had sensitivity, specificity, negative predictive value and positive predictive value of 0.81, 0.87, 0.92, 0.72 for LVO and 0.87, 0.97, 0.97, 0.86 for MT, respectively. The 10-feature RF model was superior at predicting LVO to all previous stroke scales (AUC 0.944 vs 0.759–0.878) and the 8-feature RF model was superior at predicting MT (AUC 0.970 vs 0.746–0.834). ConclusionRandom forest machine learning models utilizing clinical and demographic variables predicts LVO presence and MT candidacy with a high degree of accuracy in an ischemic stroke cohort. Further validation of this strategy for triage of stroke patients requires prospective and external validation.

Prehospital identification of large vessel occlusion using the FAST-ED score.

The prehospital identification of stroke patients with large vessel occlusion (LVO) enables appropriate hospital selection and reduces the onset-to-treatment time. The aim of this study was to investigate whether the Field Assessment Stroke Triage for Emergency Destination (FAST-ED) scale could be reconstructed from existing prehospital patient reports and to compare its performance with neurologist's clinical judgement using the same prehospital data. All patients transported by ambulance using stroke code on a six-month period were registered for the study. The prehospital patient reports were retrospectively evaluated using the FAST-ED scale by two investigators. The performance of FAST-ED score (≥4 points) in LVO identification was compared to neurologist's clinical judgement ('LVO or not'). The presence of LVO was verified using computed tomography angiography imaging. A total of 610 FAST-ED scores were obtained. The FAST-ED had a sensitivity of 57.8%, specificity of 87.2%, positive predictive value (PPV) of 37.3%, negative predictive value (NPV) of 93.4% and area under curve (AUC) of 0.724. Interclass correlation coefficient for both raters over the entire range of FAST-ED was 0.92 (0.88-0.94). The neurologist's clinical judgement raised sensitivity to 79.4%, NPV to 97.1% and PPV to 45.0% with an AUC of 0.837 (p<.05). The existing patient report data could be feasibly used to reconstruct FAST-ED scores to identify LVO. The binary FAST-ED score had a moderate sensitivity and good specificity for prehospital LVO identification. However, the FAST-ED was surpassed by neurologist's clinical judgement which further increased the sensitivity of identification.

Prehospital Identification of Large Vessel Occlusions Using Modified National Institutes of Health Stroke Scale: A Pilot Study.

Stroke identification is a key step in acute ischemic stroke management. Our objectives were to prospectively examine the agreement between prehospital and hospital Modified National Institutes of Health Stroke Scale (mNIHSS) assessments as well as assess the prehospital performance characteristics of the mNIHSS for identification of large vessel occlusion strokes.Method: In this prospective cohort study conducted over a 20-month period (11/2016–6/2018), we trained 40 prehospital providers (paramedics) in Emergency Neurological Life Support (ENLS) curriculum and in mNIHSS. English-speaking patients aged 18 and above transported for an acute neurological deficit were included. Using unique identifiers, we linked the prehospital assessment records to the hospital record. We calculated the agreement between prehospital and hospital mNIHSS scores using the Bland-Altman analysis and the sensitivity and specificity of the prehospital mNIHSS.Results: Of the 31 patients, the mean difference (prehospital mNIHSS—hospital mNIHSS) was 2.4, 95% limits of agreement (−5.2 to 10.0); 10 patients (32%) met our a priori imaging definition of large vessel occlusion and the sensitivity of mNIHSS ≥ 8 was 6/10 or 0.60 (95% CI: 0.26–0.88) and the specificity was 13/21 or 0.62 (95% CI: 0.38–0.82), respectively.Conclusions: We were able to train prehospital providers to use the prehospital mNIHSS. Prehospital and hospital mNIHSS had a reasonable level of agreement and and the scale was able to predict large vessel occlusions with moderate sensitivity.

Abstract P506: Machine Learning Models Improve Prediction of Large Vessel Occlusion and Mechanical Thrombectomy Candidacy in Acute Ischemic Stroke

Background and Purpose: Early identification of large vessel occlusions (LVO) and timely recanalization are paramount in improved clinical outcomes in acute ischemic stroke. Multiple simple stroke scales have good sensitivity, but compromise specificity for predicting LVO. No scale has been shown to predict mechanical thrombectomy (MT) candidacy. Machine learning techniques are being used for predictive modeling in many aspects of stroke care and may have potential in predicting LVO presence and MT candidacy. Methods: 287 acute ischemic stroke patients from July 2018 to July 2019 at Loyola University Medical Center were included. 36 clinical and demographic variables were analyzed using machine learning and statistical algorithms, including logistic regression, extreme gradient boosting, random forest, and decision trees to build models predictive of LVO and MT. The best performing model was compared with prior stroke scales. Results: Random forest based model resulted in the highest classification performance to predict both LVO and MT outcomes with an area under the curve (AUC) of 0.90±0.07 and 0.94±0.04, respectively. When the predictors were reduced to 7, random forest maintained a high AUC for predicting LVO (0.89). When reduced to 10 predictors, the random forest model predicted MT with an AUC = 0.93. Random forest models had excellent sensitivity and specificity of 0.86 and 0.89 for LVO and 0.89 and 0.95 for MT, respectively. The negative predictive value was 0.94 for LVO and 0.98 for MT while the positive predictive value was 0.77 for LVO and 0.79 for MT. With equal sensitivity, the random forest model was favorable to all previous stroke scales. Conclusion: Machine learning utilizing clinical and demographic variables predicts LVO and patient candidacy for MT with a high degree of accuracy. Further validation of this strategy for triage of stroke patients requires prospective and external validation.

Abstract P355: Real-Word Performance of Two Automated Software Platforms for Identification of Salvageable Tissue in Stroke Patients: A Single Center Experience

Background and purpose: Modern stroke treatment has been revolutionized by image-guided selection of patients for endovascular thrombectomy. Current automated platforms allow for real-time identification of large vessel occlusion and salvageable brain tissue. We sought to evaluate the performance of these platforms with regard to identification of infarcted and salvageable tissue. Methods: We studied all patients that presented to Henry Ford Health System hospitals over a period of 6 weeks, received CT perfusion imaging of the brain upon initial presentation. The images were processed with two automated software platforms. We prospectively measured volumes of tissue with cerebral blood flow (CBF) &lt; 30% of contralateral hemisphere, Tmax &gt;6 secs, and hypoperfusion indices (defined as the ratio of volumes Tmax&gt;10 secs and Tmax&gt;6 secs). We compared the outputs of the two platforms and analyzed the performance of each platform. Results: 66 scans were included in our study. Both platforms were able to image all stroke patients within their FDA-approved indications. With regard to all scans, both platforms were noted to demonstrate comparable CBF&lt;30% volumes (6.32 ml. vs 4.97 ml, p=0.276), and hypoperfusion indices (0.278 vs 0.338, p=0.344). However, there was statistically significant discrepancy in the volumes of tissue with Tmax&gt;6 secs (23.96 vs 14.18 ml, p=0.023). Analysis of a subset of 12 scans, with evidence of LVO or severe symptomatic stenosis on corresponding CTA, showed again comparable CBF&lt;30% volumes (12.84 ml vs 13.67 ml, p=0.725), and hypoperfusion indices (0.344 vs 0.314, p=0.699). However, the Tmax&gt;6 secs volume discrepancy was greater and still statistically significant (75.54 ml vs 39.58 ml, p=0.048) Conclusions: Automated software platforms are an invaluable aid in the identification of salvageable tissue, and selection of patients for endovascular thrombectomy in the 6-24 hour window. However, the substantial difference in the identified volumes of hypoperfused tissue-at-risk may result in largely different clinical decisions and patient outcomes. Further validation efforts (and harmonization of algorithms) are required. Stroke teams should be aware of the limitations of automated analysis and need for expert review.

Abstract P337: RAPID-LVO for Automated Detection of Intracranial Large Vessel Occlusion in Ct Angiography of the Brain

Background: Identification of large vessel occlusion (LVO) is paramount in the urgent evaluation of acute ischemic stroke (AIS). Emergent interpretation of large and high-complexity data sets, however, may impose strains upon imaging and clinical workflows, motivating development of fast and accurate computer-aided approaches to facilitate LVO detection in the emergency setting. This study investigates the performance of a fully automated LVO detection platform in a mixed cohort of stroke subjects with and without LVO on head and neck CT angiography (CTA). Methods: CTA from two cerebrovascular trials were enriched with cases from eleven global sites. Imaging and clinical variables were balanced between populations including in LVO positivity and across demographic and imaging environments to the extent achievable. Independent and fully blinded review for intracranial ICA or MCA M1 LVO was performed by two subspecialty neuroradiologists. A novel, user-independent imaging analysis application ( RAPID-LVO , iSchemaview inc) was used to predict LVO presence, location, and overall performance relative to reader consensus. Any discordance between readers was adjudicated by a blinded tertiary reader with subspecialty training. Sensitivity, specificity, and receiver-operating characteristics were determined by an independent statistician. Performance thresholds were set a priori, including a lower bound of the 95% CI of sensitivity and specificity of ≥0.8 at mean times-to-notification &lt;3.5 minutes. Results: 217 CTA (median age 65.5, 53% male, 109 LVO(+)) were included. Lower confidence limits of sensitivity and specificity exceeded 90% (sensitivity 0.963, 95% CI 0.909-0.986; specificity 0.981, 95% CI 0.935-0.995), surpassing pre-specified performance benchmarks. Subgroup analyses revealed no decrement in performance relative to subject age or sex, vendor systems, or location of the examination within or outside the United States. The area under the receiver operating characteristics curve was 0.99 (95% CI: 0.971-0.999) and average time-to-notification was 3.18 minutes. Conclusion: RAPID-LVO offers fast, highly accurate, and fully user-independent large vessel occlusion detection across all tested clinical and imaging environments.

Abstract P377: Real-Word Performance of Two Automated Software Platforms for Large-Vessel Occlusion Identification in Acute Ischemic Stroke Patients: A Single Center Experience

Background and purpose: Modern stroke treatment has been revolutionized by image-guided selection of patients for endovascular thrombectomy. Current automated platforms allow for real-time identification of large vessel occlusion and salvageable brain tissue. We had previously demonstrated that a “CTA for all” policy for stroke patients immediately upon arrival assists in the earlier identification of treatment candidates. We now sought to evaluate the performance of these platforms under this policy. Methods: All patients that presented to Henry Ford Health System hospitals over a period of 6 weeks received CTA of the head and neck upon initial presentation. The images were processed with two automated software platforms. We prospectively collected processing times, large-vessel-occlusion alerts, performance warnings, and LVO density ratios. We compared these with the interpretations of board-certified radiologists, and analyzed the performance of each platform. Results: 276 patients presented with stroke symptoms and received CT angiography upon presentation. Both platforms were able to image all stroke patients within their FDA-approved indications. Both platforms were noted to have comparable sensitivity, specificity, PPV and NPV, and excellent accuracy. The overall prevalence of LVO was extremely low (8/276). As a result, for both, NPV was much better than PPV because of the percentage of false positive results. Further ROC analysis, demonstrated an area under the ROC curve of 0.982, and overall model quality of 0.97. Optimal LVO density cutoff was &lt;0.093 in order to maximize overall accuracy, or &lt; 0.271, in order to maintain a sensitivity of 100% as an absolute priority (both significantly lower than the current threshold of &lt;0.45). Conclusions: Automated software platforms are an invaluable aid in the selection of patients for endovascular thrombectomy. Different LVO detection algorithmic thresholds may be necessary (and should be part of individual stroke center validation pathways) to avoid fatigue alert, and optimize test accuracy, when LVO prevalence is low. Stroke teams should be aware of the limitations of automated analysis and need for expert review.

FAST-ED scale smartphone app-based prediction of large vessel occlusion in suspected stroke by emergency medical service.

Background and Purpose:Considering the highly time-dependent therapeutic effect of endovascular treatment in patients with large vessel occlusion–associated acute ischemic stroke, prehospital identification of large vessel occlusion and subsequent triage for direct transport to a comprehensive stroke center offers an intriguing option for optimizing patient pathways.Methods:This prospective in-field validation study included 200 patients with suspected acute ischemic stroke who were admitted by emergency medical service to a comprehensive stroke center. Ambulances were equipped with smartphones running an app-based Field Assessment Stroke Triage for Emergency Destination scale for transmission prior to admission. The primary measure was the predictive accuracy of the transmitted Field Assessment Stroke Triage for Emergency Destination for large vessel occlusion and the secondary measure the predictive accuracy for endovascular treatment.Results:A Field Assessment Stroke Triage for Emergency Destination ⩾4 revealed very good accuracy to detect large vessel occlusion–related acute ischemic stroke with a sensitivity of 82.4% (95% confidence interval = 65.5–93.2), specificity of 78.3% (95% confidence interval = 71.3–84.3), and an area under the curve c-statistics of 0.89 (95% confidence interval = 0.85–0.94). Field Assessment Stroke Triage for Emergency Destination ⩾4 correctly identified 84% of patients who received endovascular treatment [73.5% specificity (95% confidence interval = 66.4–79.8)] with an area under the curve c-statistics of 0.82 (95% confidence interval = 0.74–0.89). In a hypothetical triage model of an urban setting, one secondary transportation would be avoided with every fifth patient screened.Conclusion:A smartphone app-based stroke triage completed by emergency medical service personnel showed adequate quality for the Field Assessment Stroke Triage for Emergency Destination to identify large vessel occlusion–associated acute ischemic stroke. We demonstrate feasibility of the use of a medical messaging service in prehospital stroke care. Based on these first results, a randomized trial evaluating the clinical benefit of such a triage system in an urban setting is currently in preparation.Clinical Trial Registration: https://clinicaltrials.gov Unique identifier: NCT04404504.

Correlation between CT angiography and digital subtraction angiography in acute ischemic strokes

ObjectiveCT angiography (CTA) is not necessarily performed for all acute ischemic strokes due to variations in national guidelines across different regions. It follows that in the absence of CTA, missed identification of large vessel occlusion (LVO) potentially leads to opportunity lost for endovascular thrombectomy. Although the accuracy of CTA is well validated in chronic arterial stenosis, it has not been adequately investigated in acute ischemic stroke. We aimed to investigate the accuracy of CTA compared with digital subtraction angiography (DSA) in detecting LVO in acute ischemic stroke. MethodsThis was a retrospective study of acute ischemic strokes with large vessel occlusion which underwent endovascular thrombectomy. We included patients who had a CTA prior to DSA and did not receive intravenous thrombolysis. Images were reviewed by 2 blinded assessors. Positive predictive value (PPV), and negative predictive value (NPV) of CTA were calculated against DSA. ResultsSeventy-seven patients were included. The median age was 67 (IQR 57−78) and 46 (59.7 %) were male. Median NIHSS was 18 (IQR 12−22). There were 284 arterial segments categorized into 215 anterior arterial segments in 54 patients and 69 posterior arterial segments in 23 patients. The median time between CTA and DSA was 126 min (IQR 91−153 min). CTA showed PPV of 91.1 % and NPV of 95.1 % compared with DSA. ConclusionsWe showed that CTA was reasonably accurate in identifying large vessel occlusion in acute ischemic stroke. We propose that current regional guidelines should include CTA for all acute ischemic strokes.

Prehospital Triage of Patients with Suspected Stroke (PRESTO): A Prospective In-Field Validation of Eight Prehospital Stroke Scales to Detect Intracranial Large Vessel Occlusion

Background: Due to the time-sensitive effect of endovascular treatment (EVT), rapid prehospital identification of large vessel occlusion (LVO) in suspected stroke patients is essential to optimize outcome. Inter-hospital transfers are an important cause of delay of EVT. Prehospital stroke scales have been proposed to select patients with LVO for direct transport to an endovascular-capable intervention centre. We aimed to prospectively validate eight prehospital stroke scales in the field. Methods: In a multi-centre prospective observational cohort study, we included suspected stroke patients transported by ambulance in the Southwest Netherlands. Paramedics assessed items from eight prehospital stroke scales: Rapid Arterial oCclusion Evaluation (RACE), Los Angeles Motor Scale (LAMS), Cincinnati Stroke Triage Assessment Tool (C-STAT), Gaze-Face-Arm-Speech-Time (G-FAST), Prehospital Acute Stroke Severity (PASS), Cincinnati Prehospital Stroke Scale (CPSS), Conveniently-Grasped Field Assessment Stroke Triage (CG-FAST), and the FAST PLUS Test. Patients that presented beyond six hours from symptom onset were excluded. Primary outcome was the clinical diagnosis of ischaemic stroke with a proximal intracranial LVO in the anterior circulation (aLVO) on CT-angiography. Prehospital stroke scale performance was expressed as the area under the receiver operating characteristic curve (AUC) and was compared with National Institute of Health Stroke Scale (NIHSS) scores assessed by physicians at the emergency department. Findings: We included 1039 patients, of which 121 patients (12%) were diagnosed with aLVO. AUCs for the prehospital stroke scales ranged from 0·69 (FAST PLUS) to 0·82 (RACE). The performance of RACE was not significantly different from G-FAST (AUC=0·80, p=0·14) and CG-FAST (AUC=0·80, p=0·09), but was significantly different from the other scales (p-values<0·05). The NIHSS performed somewhat better (AUC 0·86, p-values<0·05). Interpretation: Prehospital stroke scales detect aLVO with good accuracy and approach the performance of the physician-assessed NIHSS. RACE is the best performing prehospital stroke scale, but does not differ significantly from G-FAST and CG-FAST. Trial Registration: This study was registered at the Netherlands Trial Register (www.trialregister.nl), NL7387 Funding: BeterKeten collaboration and Theia Foundation (Zilveren Kruis) Declaration of Interests: Diederik Dippel and Aad van der Lugt report funding from the Dutch Heart Foundation, Brain Foundation Netherlands, The Netherlands Organisation for Health Research and Development, Health Holland Top Sector Life Sciences & Health, and unrestricted grants from Penumbra Inc., Stryker, Stryker European Operations BV, Medtronic, Thrombolytic Science, LLC and Cerenovus for research, all paid to institution. Pieter Jan van Doormaal reports funding from Stryker, paid to institution and an unrestricted fee from Bayer. All other authors declare no conflict of interest. Ethics Approval Statement: This study was conducted in accordance with the Dutch Agreement on Medical Treatment Act and the European General Data Protection Regulation. The Institutional Review Board of the Erasmus MC University Medical Centre has reviewed the study protocol and confirmed that the Dutch Medical Research Involving Human Subjects Act is not applicable. Because our study met the exceptions of informed consent regulations, the need for informed consent was waived. Patients or their relatives were informed about the study and could withdraw from the study through an opt-out system.

Abstract 99: Mobile Stroke Unit CTA and Direct Notification of Interventional Team Shortens Door-to-Puncture Time by One Hour

Introduction: Endovascular thrombectomy (ET) is an effective but time sensitive treatment of acute ischemic stroke. Time from Emergency Department (ED) arrival to start of ET (door-to-puncture time, DTPT) is a modifiable metric. One of the most time consuming steps in prolonging DTPT is identification of large vessel occlusion (LVO) by CT angiography (CTA). BEST-MSU is a prospective multicenter comparative effectiveness study of tPA-eligible patients managed on a mobile stroke unit (MSU) vs Emergency Medical Services (Standard Management, SM). After discovering that DTPT was greater than 60 minutes in both groups at three BEST-MSU centers in 2018, we began to routinely obtain CTA on the MSU and directly alert the ET team at receiving hospitals if a LVO was identified. We hypothesized this would shorten DTPT by over 30 minutes. Methods: In this single center experience, we compared the median (interquartile range, IQR) DTPT and MSU on-scene time for MSU patients having on-board CTA and then ET from 9/2018 to 7/2019 to corresponding MSU ET patients (excluding any that had on-board CTA) from 8/2014 to 8/2018. All CTAs were completed after tPA bolus and during tPA infusion on a Ceretom 8 slice scanner with OptiStat hand injector. All imaging occurred on-scene with the MSU stationary. Consent was obtained for all patients and strict radiation safety guidelines followed. Results: 13 consecutive patients having CTA on-board the MSU and then ET were compared to 84 patients in the pre-on-board CTA group. Baseline characteristics including median NIHSS score (20 in both groups) and frequency of tPA (85% on-board CTA vs 89% pre-on-board CTA) were comparable. Median DTPT was 60 minutes shorter with on-board CTA and direct notification of the interventional team from the MSU; 34 minutes (IQR 30-57) vs 94.5 minutes (IQR 69.75-117.25) (p &lt; 0.001). Despite the additional time to obtain the CTA on the MSU, on-scene time was only slightly prolonged and did not offset the reduction in DTPT (on-board CTA 30 minutes (IQR 28-33) vs pre-on-board CTA 27 minutes (IQR 23-31) (p = 0.01). Conclusion: Pre-hospital identification and notification of LVO by a MSU allows a one hour reduction of DTPT, and can be utilized to establish a direct to angiosuite protocol.

Abstract WP61: Artificial Intelligence for Automated Detection of Large Vessel Occlusion Strokes

Introduction: Identification of Large Vessel Occlusion (LVO) in acute ischemic stroke (AIS) patients is critical for proper decision-making. Limited availability of trained experts and delays in LVO recognition can have a detrimental effect on outcomes. We sought to evaluate an artificial intelligence-based algorithm for LVO detection in AIS. Methods: A retrospective analysis of a prospectively-collected database of AIS patients admitted to a large volume stroke center between 2014-2018 was performed. Experienced vascular neurologists graded CTA for presence and site of LVO. Concurrently, studies were analyzed by the Viz-LVO Algorithm® version 1.4 (GA) - a convolutional neural network programmed to detect occlusions from the internal carotid artery terminus (ICA-T) to the sylvian fissure, which would include all MCA M1-segment and most M2-segment lesions. CTA readings were categorized as LVOs (ICA-T, MCA-M1, MCA-M2) versus non-LVOs/more distal occlusions. Comparisons between human and AI-based readings were done by accuracy analysis and calculating Cohen’s kappa. Results: A total of 610 CTAs were analyzed. The AI algorithm rejected 3.4% of the CTAs due to poor quality. Viz-LVO identified LVOs with an overall sensitivity of 81.3%, specificity of 87.8%, and accuracy of 83.2% (AUC 0.845 (95%CI:0.81-0.88, p&lt;0.001). Table 1 shows the results per occlusion site. Accuracy was higher for ICA-T and M1 occlusions as compared to M2 occlusions. The mean run time of the algorithm was 2.78±0.5minutes. Conclusion: Our study demonstrates that automated AI reading allows for fast and accurate identification of LVO strokes. Future efforts should be made to improve the detection of the more distal occlusions.

Abstract WMP21: Machine Learning Identification of Large Vessel Occlusion (LVO) on Non-Contrast Computed Tomography (NCCT) Images

Purpose: To validate a Machine Learning algorithm able to identify LVO on NCCT. Methods: Patients with suspected acute stroke who underwent NCCT+CT Angiography (CTA) from two comprehensive stroke centers were included. Patients with intracranial haemorrhage were excluded. Two experienced radiologists identified the presence of LVO on CTA (NR-CTA) tagging the clot location and manually segmenting the clot. Acute ischemia and clot signs on NCCT were also depicted with assistance of the CTA clot location. With this information a deep learning system was used to create an algorithm (Deepstroke) to identify and locate the presence/absence of acute ischaemia and clot signs in NCCT. Deepstroke image output was used to train a binary classifier to determine LVO on NCCT. Cross-validation was performed in a stratified 5-fold of the data, including deep learning training. We also studied the effect on Deepstroke accuracy when adding the patients NIHSS and time from onset to the model (Deepstroke+). Results: The data cohort included 1354 patients, 724 (53%) with LVO by NR-CTA. The accuracy of Deepstroke to identify LVO had an AUC of 0.81 (sensitivity 0.85; specificity 0.49, PPV 0.66, NPV 0.74), and improved combined with NIHSS and time from symptom onset to AUC 0.88 (sensitivity 0.87, specificity 0.68, PPV 0.76, NPV 0.82). Deepstroke performed better on larger occlusions (Table). Among patients identified as LVO by Deepstroke+ only 19% showed no findings on NR-CTA. The agreement in LVO detection between NR-CTA and Deepstroke+ was 0.78 (Deepstroke was 0.68). Process time per patient was below 120s. Conclusions: In patients with suspected acute stroke, Deepstroke identified LVO in NCCT with a high correlation with radiologist readings of CTAs. Deepstroke could reduce the need to perform CTA, generate alarms and increase the efficiency of patients transfers in the acute management in stroke networks. Deepstroke accuracy will improve as more cases are added to the training set.

Abstract WP76: Artificial Intelligence CTA Vessel Density Mapping to Enhance Identification of Large Vessel Occlusions in Mobile Stroke Unit

Background: Equipped with CT scanners capable of imaging the brain parenchyma and vasculature, Mobile Stroke Units (MSU) have the ability to image, diagnose and treat stroke patients in the prehospital setting. Automated CTA vessel density mapping could enhance frontline neurologist scan review in identifying large vessel occlusion (LVO), ensuring appropriate patient diagnosis and routing. Methods: We analyzed consecutive acute ischemic stroke patients undergoing CTA imaging in a regional Mobile Stroke Unit. Automated CTA vessel density mapping was performed in the field immediately after scan completion. CTA source images were wirelessly transferred to an off-site processing server (RAPID.Ai, IschemiaView) for artery reconstruction and color-coded density mapping, with blue, green, yellow, and red color shading indicating vessel density decreases of 70%-85%, 60%-75%, 45%-60%, and &lt;45%. Results: Among all 16 patients, median processing time was 186 secs, and all images were available in time to aid clinical decision-making. Overall, automated processing yielded evaluable images in 94% (suboptimal contrast opacification precluded analysis of 1). Of the 15 diagnostically adequate exams, 100% (15/15) showed concordance for identification of anterior circulation occluded/abnormal vessel territories between automated CTA vessel density mapping and expert physician final CTA interpretation. Cases included true positives in 7, and true negatives in 8. Among true positives, CTA vessel density mapping identified the symptomatic occlusion in 6/6 and also correctly identified a severe cervical ICA stenosis unrelated to the clinical presentation in 1/1. Correctly detected intracranial occlusions included: ICA-17%. M1-17%, M1-M2 junction-17%, and M2-50%. Degree of vessel density diminution correlated with proximal-distal occlusion location. Conclusion: CTA vessel density mapping can feasibly and efficiently be conducted in Mobile Stroke Units and shows high accuracy in detection of large and medium intracranial vessel occlusions. Extension of mapping to the intracranial posterior circulation and algorithmic adjustment for proximal cervical stenoses/occlusions would further improve utility in aiding prehospital routing.

Rapid Identification of Large Vessel Occlusion Stroke Subtype in the PreHospital Setting

Background and Hypothesis: Stroke treatment is highly time-sensitive, with an estimated 1.9 million neurons dying per minute during an untreated ischemic stroke. The recent advent of mechanical thrombectomy (MT) and its illustrated safety and efficacy in treating large vessel occlusion (LVO) strokes has generated a need to rapidly identify LVO patients who may otherwise be brought to the nearest hospital, which may not have the capability to perform the procedure. Accurate identification of LVO in the pre-hospital setting would allow immediate EMS transport to an MTcapable Comprehensive Stroke Center, thus reducing time-to-treatment and improving patient outcome. While various grading scales, such as the C-STAT, have been developed for this purpose, all have shown to lack sensitivity and specificity for accurate LVO determination. We hypothesize that a new scale combining common LVO presentations as positive values and those of other stroke subtypes, such as small vessel occlusion (SVO) and cardioembolic stroke (CE), as negative values will increase the accuracy of LVO determination.&#x0D; Methods: This is a retrospective chart review analysis of 86 patients evaluated for stroke between January 2017-May 2018 at the Parkview Regional Medical Center with imaging confirmed LVO, SVO or CE diagnoses. &#x0D; Results: C-STAT stroke scale had a sensitivity of 54.5% and a specificity of 86.7% in differentiating LVO from other stroke subtypes. Compared to C-STAT, our new model showed a significantly higher sensitivity of 81.8% (p=0.0038) and a nonsignificant decreased specificity of 75.0% (p=0.061). &#x0D; Conclusion: Our findings suggest that our new scale combining common clinical presentations in LVO stroke patients as positive predictor values and those in SVO and CE stroke patients as negative predictor values may allow for a more accurate determination of LVO stroke in the pre-hospital setting without significant delay. A prospective, larger patient cohort in a pre-hospital setting is needed to validate these findings.

Effect of Known Atrial Fibrillation and Anticoagulation Status on the Prehospital Identification of Large Vessel Occlusion

Introduction: The Cincinnati Prehospital Stroke Severity scale (C-STAT), Los Angeles Motor Scale (LAMS), Rapid Arterial Occlusion Evaluation (RACE) score, and Field Assessment Stroke Triage for Emergency Destination (FAST-ED) scales were designed to aid emergency responder identification of patients with stroke related to large vessel occlusion (LVO). We hypothesized that the addition of a known history of atrial fibrillation (AF) without anticoagulation to currently used scales would improve LVO detection. Methods: Medical records of patients admitted to a Comprehensive Stroke Center with acute ischemic stroke in 2014-2015 were reviewed. LVO identification using the C-STAT, LAMS, RACE, and FAST-ED scores and the AF variable were compared using univariable analyses. The areas under the receiver operating curves (AUCs) were then compared for each score, the AF variable, and each score with the addition of the AF variable. Results: The sample included 233 patients without and 188 patients with an LVO. A history of known AF, history of AF with no anticoagulation, and the C-STAT, LAMS, RACE, and FAST-ED scores were each associated with LVO in univariable analyses. The AUCs for C-STAT, LAMS, RACE, and FAST-ED were similar. The addition of the known AF and no anticoagulation variable did not appreciably change these AUCs. Conclusion: Although known AF with no anticoagulation was associated with LVO in patients with acute ischemic stroke, this historic feature did not improve the accuracy of existing LVO detection scales.

Cincinnati Prehospital Stroke Scale for EMS Redirection of Large Vessel Occlusion Stroke.

Prehospital identification of large vessel occlusion (LVO) stroke may expedite treatment by direct transport to comprehensive stroke centers (CSCs) with endovascular capabilities. The Cincinnati Prehospital Stroke Scale (CPSS) is commonly used for prehospital stroke detection. We aimed to assess whether (1) a high CPSS score can identify LVO and (2) an Emergency Medical Service (EMS) redirection protocol based on high CPSS accelerated endovascular treatment (EVT). A retrospective comparison of patients transported by EMSs for suspected stroke to a high-volume CSC over a 16-month period, before and after implementation of an EMS redirection protocol based on high CPSS score (3/3). Charts were reviewed to determine the presence of LVO. Time to EVT and 3-month outcomes were compared before and after implementation. A prehospital CPSS 3/3 score was found in 223 (59%) patients, demonstrating positive and negative predictive values for LVO of 29% and 94%, respectively. CPSS-based EMS redirection increased the proportion of EVT performed after direct transport to CSC [before: 21 (36%), after: 45 (63%), p < 0.01] and decreased median first door-to-groin puncture time by 28 minutes [109 (interquartile range (IQR) 64-116) versus 81 (IQR 56-130), p = 0.03]. At 3 months, the proportion of patients achieving functional independence (modified Rankin score 0-2) went from 20/57 (35%) to 29/68 (43%) (p = 0.39) following implementation. CPSS-based EMS redirection accelerated identification of LVO strokes in the out-of-hospital setting and decreased time to EVT. Nevertheless, this protocol was also associated with high rates of non-LVO stroke. Impact on clinical outcomes should be evaluated in a larger cohort.

Initial Emergency Room Triage of Acute Ischemic Stroke.

Early recognition and differentiation of acute ischemic stroke from intracranial hemorrhage and stroke mimics and the identification of large vessel occlusion (LVO) are critical to the appropriate management of stroke patients. In this review, we discuss the current evidence and practices surrounding safe and efficient triage in the emergency room. As the indications of stroke intervention are evolving to further improve stroke care, focus has begun to revolve around recognition of LVO and provision of endovascular thrombectomy with or without the administration of tissue plasminogen activator. Systems of stroke care are being organized to achieve this goal without delay. Clinical history is important in determining time of onset or last known well time, but, alone or along with an examination, it cannot reliably predict an LVO or exclude intracranial hemorrhage and stroke mimics. The choice of imaging is influenced mainly by the duration of symptoms. On the basis of recent trials, patients presenting after the 6-h therapeutic window can be considered for endovascular thrombectomy if the computed tomographic or magnetic resonance perfusion imaging shows favorable findings. The Society of NeuroInterventional Surgery has established time metrics for each step of triage and initial management. Hospitals are required to develop multidisciplinary stroke teams and emergency protocols to meet these goals. There also needs to be coordination of the emergency medical services with the emergency facility of an appropriate stroke center (a primary stroke center, comprehensive stroke care center, or a thrombectomy-capable stroke center).

Improving field Identification of Large vessel occlusion: A retrospective review analysis (P2.3-045)

Improving field Identification of Large vessel occlusion: A retrospective review analysis (P2.3-045)

Performance of the RACE Score for the Prehospital Identification of Large Vessel Occlusion Stroke in a Suburban/Rural EMS Service

Introduction: Emergency Medical Services (EMS) providers may identify and preferentially transport patients experiencing large vessel occlusion (LVO) stroke to appropriate treatment centers. The Rapid Arterial oCclusion Evaluation (RACE) scale was created for prehospital LVO detection, yet few studies have evaluated its function in real-world EMS settings. Our objective was to assess the prehospital performance of the RACE scale for detecting LVO stroke following implementation at a large suburban/rural agency in the United States. Methods: In this retrospective analysis, all 9-1-1 patients with an EMS provider primary or secondary impression of stroke treated by the agency between June 1, 2016 and November 1, 2017 were eligible for inclusion. Patient data were abstracted using a standardized form completed by receiving hospitals. Performance for LVO detection at each RACE cutoff value was evaluated using sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV). A receiver operating characteristic (ROC) curve was used to evaluate the discrimination of the RACE scale overall. A secondary analysis of RACE for patients experiencing strokes best treated at comprehensive stroke centers (LVO and intracerebral hemorrhage [ICH]) was conducted. Results: There were 440 patients with a documented RACE score and hospital outcome data included in the analysis. About half (51%, n = 220) were female and the median age was 70 years (IQR: 59–81). Last known well time was under 4.5 hours for 76% of patients (n = 261). Over half (61%, n = 269) had a hospital discharge diagnosis of stroke and 64/440 (15%) were classified as LVO. The ROC curve demonstrated adequate discrimination with a c-statistic of 0.72. Performance for identifying LVO in the prehospital setting was greatest for RACE scores ≥5 with a sensitivity of 66% and specificity of 72%, PPV of 29%, and NPV of 93%. A RACE score ≥5 for both LVO and ICH demonstrated sensitivity: 63%, specificity: 77%, PPV: 47% and NPV: 86%. Conclusion: The RACE scale demonstrated acceptable discrimination, yet the sensitivity and positive predictive value were lower in this cohort of EMS professionals in the United States than in the original validation study conducted in Spain. Further work is needed to determine the optimal prehospital screening tool for identification of LVO.