Occlusion Detection Research Articles

Analysis of Advanced Imaging Utilization in Low National Institutes of Health Stroke Scale Score and Rate of Large Vessel Occlusion Detection in Large Telestroke Database (S24.009)

<h3>Objective:</h3> Analyze the rate of large vessel occlusion (LVO) detection in low National Institutes of Health Stroke Scale (NIHSS) score patients in a large multicenter database. <h3>Background:</h3> The benefits of mechanical thrombectomy in low NIHSS score (&lt;6) is unclear and undergoing investigation. Currently, great variation exists in practice patterns of computed tomography angiography (CTA) for patients with low NIHSS. Current guidelines do not address low NIHSS largely owing to a paucity of data. Better understanding of the rate of LVO associated with low NIHSS could help guide clinical practice and pave the way for expanding guideline recommendations. <h3>Design/Methods:</h3> Acute stroke consultations seen in the emergency department in 227 facilities (27 states) from July 1, 2021 to December 31, 2021 were extracted from the TeleCare™ database. The encounters seen within 24 hours of last known normal were reviewed for CTA performed, LVO detected, and NIHSS score. Multivariate analysis was performed to determine the odds ratio (OR) for CTA performed, and LVO found at each NIHSS score of 0–5 with reference of ≥6. <h3>Results:</h3> 23,166 acute stroke encounters were included and 10,507 had CTA performed. There were lesser odds for CTA being performed among patients with an NIHSS of: 0 (OR=0.14, [0.13,0.15]); 1 (OR=0.16, [0.15,0.18]); 2 (OR=0.27, [0.24,0.3]); 3 (OR=0.33, [0.3,0.37]); 4 (OR=0.49, [0.43,0.55]); and 5 (OR=0.71, [0.61,0.82]). In addition, there were lesser odds of an LVO detected among patients with an NIHSS of: 0 (OR=0.1, [0.07,0.13]); 1 (OR=0.09, [0.06,0.12]); 2 (OR=0.16, [0.12,0.22]); 3 (OR=0.14, [0.1,0.2]); 4 (OR=0.24, [0.18,0.32]), and 5 (OR=0.27, [0.2,0.37]). <h3>Conclusions:</h3> Analysis of our large telemedicine database demonstrated a significant number of LVOs detected with NIHSS &lt;6, supporting lowering the minimum NIHSS score for emergent CTA. The cutoff of ≥4 may be optimal threshold for LVO capture. We suggest a rational approach to CTA performance to include patients with NIHSS ≥4 or cortical symptoms. <b>Disclosure:</b> Dr. Sevilis has stock in Moderna. Dr. Fowler has nothing to disclose. Dr. Avila has nothing to disclose. Ms. Boyd has nothing to disclose. Mr. Collins has nothing to disclose. Dr. Gao has nothing to disclose. Dr. Heath has received personal compensation in the range of $0-$499 for serving as a Epidemiologic analysis consultant with Neuroscience Innovation Foundation . Dr. Devlin has received personal compensation for serving as an employee of Neuroscience Innovation Foundation. Dr. Devlin has received personal compensation in the range of $10,000-$49,999 for serving as a Consultant for Nova Signal. Dr. Devlin has received personal compensation in the range of $10,000-$49,999 for serving as a Consultant for Viz.ai. Dr. Devlin has received personal compensation in the range of $500-$4,999 for serving as a Consultant for Medtronic. Dr. Devlin has received personal compensation in the range of $500-$4,999 for serving on a Speakers Bureau for Medtronic. Dr. Devlin has received personal compensation in the range of $500-$4,999 for serving on a Speakers Bureau for Viz,ai. Dr. Devlin has received personal compensation in the range of $5,000-$9,999 for serving as an officer or member of the Board of Directors for Neuroscience Innovation Foundation. Dr. Devlin has stock in Nova Signal. Dr. Devlin has stock in Viz.ai. The institution of Dr. Devlin has received research support from Viz.ai. Dr. Devlin has received research support from Nova Signal. Dr. Devlin has received intellectual property interests from a discovery or technology relating to health care.

A novel attention-based enhancement framework for face mask detection in complicated scenarios

In the context of COVID-19 pandemic prevention and control, it is of vital significance to realize accurate face mask detection via computer vision technique. In this paper, a novel attention improved Yolo (AI-Yolo) model is proposed, which can handle existing challenges in the complicated real-world scenarios with dense distribution, small-size object detection and interference of similar occlusions. In particular, a selective kernel (SK) module is set to achieve convolution domain soft attention mechanism with split, fusion and selection operations; a spatial pyramid pooling (SPP) module is applied to enhance the expression of local and global features, which enriches the receptive field information; and a feature fusion (FF) module is utilized to promote sufficient fusions of multi-scale features from each resolution branch, which adopts basic convolution operators without excessive computational complexity. In addition, the complete intersection over union (CIoU) loss function is adopted in the training stage for accurate positioning. Experiments are carried out on two challenging public face mask detection datasets, and the results demonstrate the superiority of the proposed AI-Yolo against other seven state-of-the-art object detection algorithms, which achieves the best results in terms of mean average precision and F1 score on both datasets. Furthermore, effectiveness of the meticulously designed modules in AI-Yolo is validated through extensive ablation studies. In a word, the proposed AI-Yolo is competent to accomplish face mask detection tasks under extremely complex situations with precise localization and accurate classification.

SELF-SUPERVISED LEARNING WITH RADIOLOGY REPORTS, A COMPARATIVE ANALYSIS OF STRATEGIES FOR LARGE VESSEL OCCLUSION AND BRAIN CTA IMAGES.

Scarcity of labels for medical images is a significant barrier for training representation learning approaches based on deep neural networks. This limitation is also present when using imaging data collected during routine clinical care stored in picture archiving communication systems (PACS), as these data rarely have attached the high-quality labels required for medical image computing tasks. However, medical images extracted from PACS are commonly coupled with descriptive radiology reports that contain significant information and could be leveraged to pre-train imaging models, which could serve as starting points for further task-specific fine-tuning. In this work, we perform a head-to-head comparison of three different self-supervised strategies to pre-train the same imaging model on 3D brain computed tomography angiogram (CTA) images, with large vessel occlusion (LVO) detection as the downstream task. These strategies evaluate two natural language processing (NLP) approaches, one to extract 100 explicit radiology concepts (Rad-SpatialNet) and the other to create general-purpose radiology reports embeddings (DistilBERT). In addition, we experiment with learning radiology concepts directly or by using a recent self-supervised learning approach (CLIP) that learns by ranking the distance between language and image vector embeddings. The LVO detection task was selected because it requires 3D imaging data, is clinically important, and requires the algorithm to learn outputs not explicitly stated in the radiology report. Pre-training was performed on an unlabeled dataset containing 1,542 3D CTA - reports pairs. The downstream task was tested on a labeled dataset of 402 subjects for LVO. We find that the pre-training performed with CLIP-based strategies improve the performance of the imaging model to detect LVO compared to a model trained only on the labeled data. The best performance was achieved by pre-training using the explicit radiology concepts and CLIP strategy.

Hand-over-face occlusion and distance adaptive heart rate detection based on imaging photoplethysmography and pixel distance in online learning

In online learning, imaging photoplethysmography (iPPG) can extract biometric information from the students’ face video, which can be used for further analysis of learning status. However, the iPPG-based methods cannot detect the heart rate accurately when the face-camera distance changes or the subject's face is occluded by the hand (Hand-Over-Face, HOF). This paper proposes a distance-adaptive heart rate detection model based on chroma and pixel distance, which can detect heart rate accurately with face-camera distance changing in a certain range. The influence factors of distance are verified on the region of interest (ROI) of the image in the process of iPPG signal extraction, and the relationship model is established between the pixel distance characteristic parameters and the physiological parameters through the BP neural network. The experimental results show that the RMSE of the model proposed in this paper is 1.50 bpm, and the model determination coefficient is 0.9504, which can effectively correct the heart rate detection error caused by face-camera distance changes. This paper proposes a gesture segmentation method GROI combined with gesture recognition to obtain an effective face ROI when HOF occurs. The experimental results show that when the ROI is occluded severely by the gesture, the RMSE for heart rate detection is reduced by 1.11 bpm, which demonstrates the effectiveness of the algorithm.

Evaluation of Reliability and Validity of Occlusal Caries Detection by Direct Visual, Indirect Visual and Fluorescence Camera Using ICDAS II (Codes 0, 1, and 2): An In Vivo Study.

Early detection of carious lesions paves the way for the preservation of tooth structures by remineralization strategies. Visual examination using international caries detection and assessment system (ICDAS) II has been validated with the gold standard of histological classification. A fluorescence camera is yet another diagnostic aid for the early detection of carious lesions based on the autofluorescence of the enamel and dentin. This study was done to evaluate the validity of the fluorescence camera (Soprocare and Acteone) in detecting early carious lesions. A total of 690 occlusal surfaces of non-cavitated premolars and molars were examined by examiners 1 and 2, trained in the ICDAS II grading system. The photographs of the occlusal surface and the fluorescence images, which were taken using a Soprocare (Acetone) intraoral camera, were examined by examiners 3 and 4. The scoring was tabulated and correlated. The κ values for interexaminer reproducibility of indirect visual was 0.841 (good), and his fluorescence camera was 1.00, which is very good. The correlation analysis revealed that there was a higher correlation between direct visual and indirect visual for both examiners when compared to direct and fluorescence cameras. There was a positive relationship between indirect visual and fluorescence cameras for both the experimenter). The indirect visual method for the detection of carries has high sensitivity and specificity irrespective of the examiner. Examination by fluorescence camera has a low sensitivity and high specificity. The specificity of the caries detection method by indirect visual examination based on ICDAS II coding that of the Soprocare fluorescence camera was consistent and reliable, whereas indirect visual examination had a high sensitivity for detecting ICDAS codes 1 and 2. Soprocare showed a very low sensitivity in detecting ICDAS code 1 and 2 lesions. Dhanavel C, Sai CK, Neelamurthy PS, et al. Evaluation of Reliability and Validity of Occlusal Caries Detection by Direct Visual, Indirect Visual and Fluorescence Camera Using ICDAS II (Codes 0, 1, and 2): An In Vivo Study. Int J Clin Pediatr Dent 2023;16(1):74-78.

Number Needing Review: A Novel Metric to Assess Triage Efficiency of Large Vessel Occlusion Detection Systems

Background Endovascular thrombectomy is the gold‐standard treatment for large vessel occlusions (LVOs). A novel metric is introduced: the number needing review (NNR) to assess the triage efficiency of LVO detection systems. Methods Patients with suspected ischemic stroke and images processed by RapidAI LVO detection software over 6 months were reviewed. Only patients with LVOs of the M1 segment were included. The NNR was calculated for an M1 occlusion. Results Of 559 patients, M1 occlusion was detected in 42 patients (7.5%). RapidAI LVO had a sensitivity of 71%, specificity of 94%, positive predictive value of 49%, and negative predictive value of 92% for M1 occlusion. When gaze deviation and hyperdense sign were combined with RapidAI LVO, the specificity and positive predictive value increased to 100% for an M1 occlusion. A negative RapidAI LVO result combined with a low (&lt;15 mL, T max &gt;6 seconds) or high (&lt;50 mL, T max &gt;6 seconds) T max threshold was found to have a specificity and positive predictive value of 100% for no occlusion. The combination of gaze deviation, hyperdense sign, positive RapidAI LVO, and negative RapidAI LVO with low T max threshold yielded an NNR of 24 per 100 cases. When combined with a negative RapidAI LVO and a high T max threshold, the NNR was 16 per 100 cases. Adding National Institutes of Health Stroke Scale score &lt;7 decreased the NNR to 9 per 100 cases. Conclusion Adding gaze deviation and hyperdense sign to the RapidAI LVO increases the specificity and positive predictive value for an M1 occlusion. When combined with a negative RapidAI LVO detection and either a low or high T max &gt;6 seconds threshold, the NNR is significantly reduced.

Development of a Machine Learning Algorithm to Automate Occlusion Detection in Lower Extremity Conventional Angiograms

Development of a Machine Learning Algorithm to Automate Occlusion Detection in Lower Extremity Conventional Angiograms

Multi object tracking in soccer video focusing on occlusion detection and resolving

Multi object tracking in soccer video focusing on occlusion detection and resolving

Abstract Number ‐ 51: Automated Versus Human Hyperdense Vessel Sign Detection Using Non‐Contrast Computed Tomography Scans

Introduction Rapid detection of large vessel occlusion (LVO) is very crucial in triaging stroke patients potentially candidates for mechanical thrombectomy (MT). Hyperdense vessel sign (HDVS) is one of the earliest ischemic changes in non‐contrast CT scan (NCCT) indicating LVO stroke. Artificial intelligence emerged to detect HDVS with the advantages of faster acquisition, less variation, and a lower need for experience than the usual detection. We aimed to identify the diagnostic performance of automated software (e‐Stroke, Brainomix) in HDVS detection. Methods A prospectively collectedMT database from March 2020 to August 2021 was reviewed. Patients were included if they had intracranial internal carotid artery or middle cerebral artery M1 or M2 occlusion. Cases with HDVS were identified through the routine 2.5‐mm slice thickness NCCT scans after being correlated with patients’ clinical information and confirmed with CT angiography (CTA) scans. NCCT scans were classified according to slice thickness into two groups: 2.5‐mm scans and 0.625‐mm generated scans. All NCCT scans were read by e‐Stroke software, then deidentified and reviewed by two stroke neurologists who were blinded to any clinical, other imaging, or therapeutic information. They were required to record the presence/laterality of HDVS before and after observing other NCCT early ischemic changes like gaze deviation, loss of insular ribbon, caudate or lentiform hypodensity. ROC curve analysis was used to estimate sensitivity and specificity and the area under the curve (AUC) was compared using DeLong’s test. Inter‐rater agreement between the two readers’ final reads, e‐Stroke, and the standard read was measured using the Fleiss Kappa test. Results Among 304 patients included in the study, 37.7% had HDVS. Approximately 44% of the scans had 2.5‐mm slice thickness and 56% had 0.625‐mm slice thickness. The e‐Stroke software identified HDVS with a sensitivity of 63% and a specificity of 71% (Table 1). The mean AUC value of e‐Stroke HDVS detection (0.67[0.61‐0.74]) was similar to reader‐1 (0.68[0.62‐0.74];p = 0.87) and reader‐2 (0.63[0.57‐0.70];p = 0.56). HDVS detection improved by reader‐1(0.78[0.72‐0.83];p = 0.03) after observing other early ischemic changes on the same scans, but reader‐2 performance remained similar to e‐Stroke (0.69[0.63‐0.76];p = 0.71). AUC, sensitivity and specificity ofHDVS detection by e‐Stroke were significantly higher using 2.5‐mm compared to 0.625‐mm sliced NCCT scans (0.78[0.70‐0.86],sensitivity 70%,specificity 86%;p&lt; 0.001) vs (0.58[0.50‐0.67],sensitivity 56%,specificity 61%;p = 0.06) respectively;p = 0.01. The readers also had higher AUC values with 2.5‐mm scans but not statistically significant, (0.74[0.66‐0.83] vs 0.64[0.56‐0.73];p = 0.18) for reader‐1 and (0.68[0.59‐0.77] vs 0.57[0.48‐0.66];p = 0.23) for reader‐2. The same after the final read, (0.85[0.78‐0.92] vs 0.75[0.67‐0.82];p = 0.08) for reader‐1 and (0.73[0.65‐0.82] vs 0.67[0.58‐0.76];p = 0.43) for reader‐2. Similarly, inter‐rater agreement was higher using 2.5‐mm sliced scans, k = 0.50(0.43‐0.75) compared to0.625‐mm scans,k = 0.27(0.21‐0.33). Conclusions Artificial intelligence (e‐Stroke software) has comparable sensitivity and specificity to human readers in HDVS detection. For e‐Stroke software, 2.5‐mm sliced CT scans are better to identifyHDVS compared to 0.625‐mm scans.

Abstract Number ‐ 205: Impact of Automated Neuroimaging Triage Platform on Time Metrics at a Thrombectomy Capable Stroke Center

Introduction Rapid detection and appropriate triage of large vessel occlusion (LVO) strokes, upon arrival to a hospital system, is key to achieving better outcomes after endovascular thrombectomy (EVT) mediated recanalization of LVO strokes . Automated neuroimaging analyses has the potential to streamline intra‐ hospital workflow. We aim to study the change in vital intra‐hospital time metrics before and after adoption of an automated neuroimaging triage platform (ANTP). Methods We performed a retrospective analysis of prospectively collected data at our EVT capable stroke center between April 2019 and November 2021. IschemaView’s Rapid software was adopted in October 2020 for triage of stroke patients. Patients treated before were our control cohort and patients treated after were our intervention cohort. Confirmed LVO strokes presenting to our center within 24 hours of stroke onset were included. Results A total of 305 patients were included‐ control (n = 150) and intervention cohort (n = 155). Age (70±15 vs 72 ±15,p = .40), NIHSS score (13 ±8 vs 14 ±7, p = .24), and last‐known‐well to arrival (392 ±464 vs 376 ±427 min, p = .38) were comparable between the two groups. Time from arrival to non‐contrast CT head was also comparable between the two groups (150 ±712 vs 161 ±794, p = .89). Time from non‐contrast CT head to decision to EVT (36 ±54 vs 22 ±58, p = .06) and time from non‐contrast CT to groin puncture (41 ±61 vs 61 ±65, p = .01) were shorter in the intervention cohort compared to control cohort. Length of stay and discharge mRS score were similar between the two cohorts. Conclusions Automated neuroimaging triage platform has the potential to improve workflow by decreasing time from non‐contrast CT to EVT decision and groin puncture. Further studies are required to study impact on patient outcomes.

Occlusal caries detection on 3D models obtained with an intraoral scanner. A validation study

ObjectivesTo evaluate the diagnostic performance of visual caries assessment on 3D dental models obtained using an intraoral scanner and to compare it with the performance of the clinical visual inspection. MethodsFifty-three permanent posterior teeth scheduled for extraction were randomly selected and included in this study. One to three independent examination sites on the occlusal surface of each tooth were clinically inspected using International Caries Detection and Assessment System (ICDAS) criteria. Afterwards, the examined teeth were scanned intraorally with a 3D intraoral scanner (TRIOS 4, 3Shape TRIOS A/S, Copenhagen, Denmark) using white and blue-violet light (415 nm wavelength) to capture the colour and fluorescence signal from the tissues. Six months after the clinical examination, the same examiner conducted the on-screen assessment of the obtained 3D digital dental models at the selected examination sites using modified ICDAS criteria. Both tooth colour and fluorescence texture with high resolution were assessed. Lastly, an independent examiner conducted the histological examination of all teeth after extraction. Using histology as the reference test, Sensitivity (SE), Specificity (SP), Accuracy (ACC), area under the Receiver Operating Characteristic (ROC) curve, and Spearman's correlation coefficient were calculated for the clinical and on-screen ICDAS assessments. ResultsThe ACC values of the evaluated methods varied between 0.59-0.79 for initial caries lesions and 0.77-0.99 for moderate-extensive caries lesions. Apart from SE values corresponding to caries in the inner half of enamel, no significant difference was observed between clinical visual inspection and on-screen assessment. In addition, no difference was found in the assessment of 3D models with tooth colour alone or supplemented with fluorescence for all the evaluated diagnostic measures. ConclusionsOn-screen visual assessment of 3D digital dental models with tooth colour or fluorescence showed a similar diagnostic performance to the clinical visual inspection when detecting and classifying occlusal caries lesions on permanent teeth. Clinical significance3D intraoral scanning can aid the detection and classification of occlusal caries as part of patient screening and can potentially be used in remote caries assessment for clinical and research purposes.

Detection of Occluded Small Commodities Based on Feature Enhancement under Super-Resolution.

As small commodity features are often few in number and easily occluded by hands, the overall detection accuracy is low, and small commodity detection is still a great challenge. Therefore, in this study, a new algorithm for occlusion detection is proposed. Firstly, a super-resolution algorithm with an outline feature extraction module is used to process the input video frames to restore high-frequency details, such as the contours and textures of the commodities. Next, residual dense networks are used for feature extraction, and the network is guided to extract commodity feature information under the effects of an attention mechanism. As small commodity features are easily ignored by the network, a new local adaptive feature enhancement module is designed to enhance the regional commodity features in the shallow feature map to enhance the expression of the small commodity feature information. Finally, a small commodity detection box is generated through the regional regression network to complete the small commodity detection task. Compared to RetinaNet, the F1-score improved by 2.6%, and the mean average precision improved by 2.45%. The experimental results reveal that the proposed method can effectively enhance the expressions of the salient features of small commodities and further improve the detection accuracy for small commodities.

The diagnostic performance of artificial intelligence algorithms for identifying M2 segment middle cerebral artery occlusions: A systematic review and meta-analysis

BackgroundArtificial intelligence (AI)-based algorithms have been developed to facilitate rapid and accurate computed tomography angiography (CTA) assessment in proximal large vessel occlusion (LVO) acute ischemic stroke, including internal carotid artery and M1 occlusions. In clinical practice, however, the detection of medium vessel occlusion (MeVO) represents an ongoing diagnostic challenge in which the added value of AI remains unclear. PurposeTo assess the diagnostic performance of AI platforms for detecting M2 occlusions. MethodsStudies that report the diagnostic performance of AI-based detection of M2 occlusions were screened, and sensitivity and specificity data were extracted using the semi-automated AutoLit software (Nested Knowledge, MN) platform. STATA (version 16 IC; Stata Corporation, College Station, Texas, USA) was used to conduct all analyses. ResultsEight studies with a low risk of bias and significant heterogeneity were included in the quantitative and qualitative synthesis. The pooled estimates of sensitivity and specificity of AI platforms for M2 occlusion detection were 64% (95% CI, 53 to 74%) and 97% (95% CI, 84 to 100%), respectively. The area under the curve (AUC) in the SROC curve was 0.79 (95% CI, 0.74 to 0.83). ConclusionThe current performance of the AI-based algorithm makes it more suitable as an adjunctive confirmatory tool rather than as an independent one for M2 occlusions. With the rapid development of such algorithms, it is anticipated that newer generations will likely perform much better.

Imaging of Central Nervous System Ischemia

This article describes imaging modalities used in the evaluation of patients presenting with symptoms of acute ischemic stroke. The year 2015 marked the beginning of a new era in acute stroke care with the widespread adoption of mechanical thrombectomy. Subsequent randomized controlled trials in 2017 and 2018 brought the stroke community even further into this new territory with the expansion of the eligibility window for thrombectomy using imaging-based patient selection, which led to an increase in the use of perfusion imaging. Now, after several years of routine use, the debate is ongoing as to when this additional imaging is truly required and when it results in unnecessary delays in time-sensitive stroke care. At this time, more than ever, a robust understanding of neuroimaging techniques, applications, and interpretation is essential for the practicing neurologist. CT-based imaging is the first step in most centers for the evaluation of patients presenting with symptoms of acute stroke because of its wide availability, speed, and safety. Noncontrast head CT alone is sufficient for IV thrombolysis decision making. CT angiography is very sensitive for the detection of large-vessel occlusion and can be used reliably to make this determination. Advanced imaging including multiphase CT angiography, CT perfusion, MRI, and MR perfusion can provide additional information useful for therapeutic decision making in specific clinical scenarios. In all cases, it is essential that neuroimaging be performed and interpreted rapidly to allow for timely reperfusion therapy.

Abstract WP143: Use Of CTP For Evaluation Of EVT Candidates In The Early Time Window (&lt;6 Hours From Lkwt) Does Not Delay EVT T And Use Of CTP Beyond 6 Hours Improves Speed Of EVT

Introduction: Computerized Tomography Perfusion (CTP) improves detection of distal occlusion and aids in patient selection for endovascular therapy (EVT) for acute ischemic stroke (AIS). As EVT continues to evolve, it is important to assess if CTP continues to aid in the evaluation of AIS or if it adds unnecessary time. Here we assess whether the imaging approach used to evaluate EVT patients was associated with improved door to device times or clinical outcomes. Specifically, we assessed whether CTA alone vs CTA/CTP combined had an effect on door to device (DTD) for EVT, discharge disposition and 90-day mRS. Methods: Data from 1357 AIS EVT patients from 17 hospitals was obtained from a multi-state health system’s stroke registry between Jan 1, 2018, to Apr 11, 2022. Multivariable mixed-effect linear and logistic regression models were employed to identify the association of evaluation type with DTD, discharge disposition and mRS, adjusting for age, sex, race, mode of arrival, NIHSS, stroke mechanism, medical history, and TICI score. Multivariable ordinal regression model was used with mRS as ordinal variable. Each clinical outcome was assessed for last known well (LKW) to arrival below 6 hours and between 6 and 24 hours. Results: The median age was 74 years, 729 (47.7%) were male; 958 (70.6%) arrived in less than 6 hours from LKW and 399 (29.4%) in the interval between 6 and 24 hours. CTA and CTP evaluation was used for 829 (61.1%) patients. In the multivariable models, evaluation with both CTA and CTP was significantly associated with a decrease of the DTD when LKW to arrival was &gt; 6 hours (n= 955, regression coefficient -16, 95% CI -27, -3.6, p = .01) and had no effect on DTD inside of 6 hours. No significant association was found between evaluation type and discharge disposition or 90-day mRs after adjustment for covariates. Conclusions: In EVT patients presenting inside of 6 hours, addition of CTP to CTA caused no delay in DTD and in those presenting from 6-24 hours adding CTP shortened DTD by 16 minutes. This may reflect that CTP allows for more confident and expeditious treatment. These results along with no effect of evaluation type on clinical outcome argue that there is no disadvantage to the use of CTP inside of 6 hours and a clear time advantage beyond 6 hours.

Abstract WP51: Effect Of Automated Large Vessel Occlusion Detection On Endovascular Therapy Treatment Rate: A Multi-Center Cluster Randomized Clinical Trial

Introduction: Automated Large Vessel Occlusion (LVO) detection and software-based care team integration may improve treatment times. Their effect on likelihood to perform Endovascular Thrombectomy (EVT) in LVO Acute Ischemic Stroke (AIS) patients remains undetermined. Methods: We performed a cluster randomized stepped wedge clinical trial across 4 comprehensive stroke centers in the greater Houston area from January 1st 2021 - February 24th 2022. This trial design was chosen due to the impracticality of randomizing at the patient level. An automated LVO detection and workflow manager software (Viz.AI) was implemented at the 4 hospitals in a stepped fashion, with a randomly determined order (Figure 1) . Patients were included if they were diagnosed with LVO AIS based on CT angiogram. Patients who presented during the 2-week transition period at the time of LVO detection software activation were excluded. The primary outcome was proportion of LVO AIS patients treated with EVT adjusted for age, sex and NIHSS and was determined using a mixed-effect logistic regression, with a random effect for cluster (hospital site) and fixed effect for exposure status. Results: Among 834 patients that met inclusion criteria, the median age was 68 [58-79] years, NIHSS 13 [5-20], and 50% were female. Median time from last known well to arrival was 298 [102- 754] minutes. A total of 575 patients were evaluated in the unexposed period and 259 in the exposed. In univariable analysis, there was no significant increase in the proportion of LVO AIS treated with EVT (40.5% vs. 45.6%, unexposed vs. exposed, p=0.17) . In multivariable adjusted mixed-effects model, EVT rates did not change due to software implementation [OR 1.18, p=0.45]. Discussion: In a multi-center cluster randomized trial, implementation of an automated LVO detection software with care team-wide communication did not significantly alter rates of EVT in patients with LVO AIS.

Abstract 27: Effect Of Automated Large Vessel Occlusion Detection On Door-in-door-out Times At Primary Stroke Centers: A Multi-center Prospective Cohort Study

Introduction: Automated LVO detection and software-based care team integration has been shown to improve door-to-groin time metrics at comprehensive stroke centers (CSCs). The impact of this software on accelerating transfers out of primary stroke centers (PSCs) for LVO AIS patients remains undetermined. Methods: We performed a prospective cohort study across 11 stroke centers (7 PSCs and 4 CSCs) in the greater Houston area from January 1st 2021 - February 24th 2022. An automated LVO detection and workflow manager software (Viz.AI) was implemented at all the hospitals. Patients were included if they were diagnosed with LVO AIS based on CT angiogram at a PSC and were transferred to CSC. Patients who presented during a 2-week transition period at the time of LVO detection software activation were excluded. The primary outcome was time from PSC arrival to departure to CSC (door-in-door-out, DIDO) before and after software implementation and was performed using the Wilcoxon rank sum test (STATA v.15). Results: Among 234 patients that met inclusion criteria, median age was 65 [58-76] years, NIHSS 3 [1-8] and 45% were female. Mean time from last known well to PSC arrival was 512 +/- 281 minutes and mean DIDO was 176 +/- 77 min. A total of 156 patients were evaluated in the pre-software implementation period and 78 after. There were no significant differences in the pre- and post-implementation cohorts with respect to age, risk factors (HTN, HLD, Afib, carotid disease, diabetes), although a greater proportion of patients in the pre-implementation phase were female (52% vs. 31%, p&lt;0.01). There was a significant reduction in the median DIDO time with software implementation (163 vs. 130 minutes, p=0.014). Discussion: In this multicenter cohort study, implementation of automated LVO detection with coupled care-team communication tool improved PSC transfer performance by over 30 minutes. These findings corroborate the findings for care acceleration at CSCs with similar software implementations.

Commentary on: diagnostic accuracy of a decision-support software for the detection of intracranial large vessel occlusion in CT angiography

Commentary on: diagnostic accuracy of a decision-support software for the detection of intracranial large vessel occlusion in CT angiography

Diagnostic accuracy of a decision-support software for the detection of intracranial large-vessel occlusion in CT angiography

To investigate the real-world clinical performance of the decision-support software "e-CTA" (e-Stroke Suite, Brainomix Limited, Oxford UK) for the detection of acute intracranial large-vessel occlusion (LVO) on computed tomography (CT) angiography at a UK district general hospital. The retrospective study included 300 consecutive CT angiograms of the head and neck performed between 8 March 2021 and 20 May 2021. e-CTA findings were recorded and compared with the radiologist report. Cases in which there was disagreement between e-CTA and the radiologist were reviewed by a sub-specialist vascular radiologist as the reference standard. The incidence of intracranial LVO was 7%. e-CTA correctly identified 18 of 21 intracranial proximal LVOs (86%). There were 34 false positives. The sensitivity was 0.86 (95% confidence interval [CI], 0.64-0.97), with specificity of 0.88 (95% CI, 0.83-0.91). The positive predictive value was 0.35 (95% CI, 0.27-0.43). The negative predictive value was 0.99 (95% CI, 0.96-1.00). Sensitivity, specificity, and negative predictive values were similar to those reported in the literature (Seker etal., Int J Stroke. 2021; 17:77-82); however, the positive predictive value for e-CTA was significantly lower. In practice, this meant that over half of all reported occlusions by the software were false positives. Radiologists should be aware of these metrics in order to assign appropriate weight to software findings when formulating a report. Differences in population demographics, scanners, CT protocols, and incidence are all factors potentially influencing software accuracy. Local validation testing may help provide accuracy metrics more relevant to individual institutions.

Face De-Occlusion With Deep Cascade Guidance Learning

Occlusion is a challenging yet commonly seen problem for facial perception. Existing works resort to deep learning models and perform model training on synthesized data due to the lack of paired real-world data. As a result, they usually perform unsatisfactorily on real-world occluded faces because of domain gaps. In this paper, we decompose the face de-occlusion task into three stages, i.e., occlusion detection, face parsing, and face reconstruction, to alleviate this issue. We first perform occlusion detection and use its results as guidance for the second stage to conduct occlusion-free face parsing. As such, face deocclusion is first performed on the face paring space with less difficulty. We can train these two stages on both synthesized and real-world images, hence can obtain accurate results for the latter. In the last stage, we use the domain-agnostic occlusion detection map and the face parsing map as the guidance to conduct face reconstruction, thus can reduce the impact of appearance information and improve the model performance on real-world data. Aiming at improving the model capacity of inferring occluded facial appearance, we also propose two types of reference modules to use relevant facial parts to enhance the reconstruction of occluded regions. Consequently, our proposed model achieves promising face de-occlusion results on real-world images.