Welcome to the ESOC 2021 Virtual Conference Calendar

Background And Aims

Delays in acute stroke treatment contribute to severe and negative impacts for patients and significant costs to the health system. Targeted approaches to optimise stroke workflow processes can improve patient outcomes but is challenging to achieve. Virtual reality (VR) provides immersive and engaging training and overcomes some existing training barriers. We recently initiated the TACTICS acute stroke trial evaluating an implementation intervention combining multimodal CT imaging and streamlined workflow training. Within this trial we developed TACTICS VR, which includes a novel portable VR training application to upskill healthcare professionals in optimal stroke workflow, user-facing website and automated analytics. TACTICS VR was developed via an extensive consultation process to ensure content was evidence-based, best-practice and tailored for the target audience.

Methods

We report implementation of TACTICS VR in 22 Australian hospitals with 177 users through March 2021. In the training, users work through a real-world stroke case and make decisions relating to stroke workflow from pre-notification through assessment, imaging interpretation and treatment.

Results

Mean training duration was approximately 21 minutes with 79% response accuracy. Survey feedback indicate a high level of usability, acceptability and utility of TACTICS VR, including aspects of hardware, software design, educational content, training feedback and implementation strategy. Further, users self-reported increased confidence in their ability to make improvements in stroke management after TACTICS VR training (post-training mean=4.1; pre-training=3.6; 1=strongly disagree, 5=strongly agree; n=58).

Conclusions

TACTICS VR is a fit-for-purpose, evidence-based training application for stroke workflow optimisation that can be readily deployed on-site in a clinical setting.

Trial Registration Number

ACTRN12619000750189

Group Name

on behalf of the MR CLEAN Registry and PRESTO investigators

Background And Aims

Machine learning algorithms hold potential to contribute to fast and accurate detection of large vessel occlusion (LVO) in patients with suspected acute ischemic stroke. We assessed the diagnostic performance of an automated LVO detection algorithm on computed tomography angiography (CTA).

Methods

Data from the MR CLEAN Registry and PRESTO were used including patients with and without LVO. CTA data were analyzed by the algorithm for detection and localization of LVO (ICA/ICA-T, M1 or M2 occlusion). Impact of scan acquisition parameters on algorithm performance was also evaluated. Assessments done by neuroradiologists were used as reference. Diagnostic performance was assessed by means of sensitivity, specificity and area under the curve (AUC).

Results

We analyzed CTA’s of 1,110 patients with LVO from the MR CLEAN Registry and of 625 patients with suspected acute ischemic stroke in PRESTO (141 with and 484 without LVO). For detection of LVO, the algorithm yielded a sensitivity of 89% in the MR CLEAN Registry and sensitivity of 72%, specificity of 78%, and AUC of 0.75 in PRESTO. Sensitivity per occlusion location was 88% for ICA/ICA-T, 94% for M1, and 72% for M2 occlusion in the MR CLEAN Registry and 80% for ICA/ICA-T, 95% for M1, and 49% for M2 occlusion in PRESTO. Slice thickness ≥2 mm reduced the AUC from 0.83 (<1mm) to 0.71 (P<0.01) and venous scan phase reduced the AUC from 0.87 (equilibrium phase) to 0.74 (P<0.01).

Conclusions

The algorithm provided a high detection rate for proximal LVO, but performance varied by occlusion location and image quality.

Trial Registration Number

trialregister.nl; NL7387

Background And Aims

Intracranial vessel perforation is a procedural complication during endovascular thrombectomy (EVT). Its occurrence is strongly associated with unfavorable treatment outcomes. Early identification of perforation would allow therapeutic actions to prevent situation deterioration. Due to its low occurrence and the large heterogeneity in image appearance, perforations may initially be missed by the interventionalists. In this work, we study the feasibility of automated intracranial vessel perforation detection during EVT using deep learning techniques.

Methods

From the MR CLEAN registry, fifty-three patients (149 acquisitions) with vessel perforations were identified and annotated by an experienced neuroradiologist. Another 150 acquisitions (from 150 patients) without perforations were randomly selected as negative samples. The proposed solution builds on top of state-of-the-art object detection algorithms. It incorporates temporal information of DSA with convolutional bidirectional gated recurrent units (GRU), further followed by a problem-tailored acquisition level optimization to reduce false positives based on temporal consistency.

Results

In ten-fold cross-validation on 203 patients (299 acquisitions, 3607 images), the proposed method achieves an area under the receiver-operating characteristic curve of 90% for acquisition-level classification of acquisitions with a perforation. The sensitivity and specificity were 81% and 86%, respectively.

Conclusions

The proposed deep learning-based algorithm achieves promising performance in vessel perforation detection in DSA for stroke patients, and can potentially be deployed in clinical practice to detect perforation early and allow direct clinical decision making.

Trial Registration Number

Not applicable

Group Name

RITeS (Real-world Independent Testing of e-ASPECTS) Collaboration

Background And Aims

Expert stroke CT interpretation is not always available. Artificial intelligence (AI) software might assist less experienced readers. We compared a medical student with software assistance against expert interpretation for intracranial haemorrhage detection and associations with outcomes.

Methods

We included all RITeS patients (from 9 stroke studies) with acute intracranial haemorrhage on baseline CT. We tested diagnostic accuracy of student aided by e-ASPECTs software (v10.0, Brainomix, UK) and agreement with masked experts for presence of intracranial haemorrhage in 3 anatomical regions (intraparenchymal, extra-axial, or intraventricular) using Cohen’s kappa, κ. We sought associations of haemorrhage location (5 regions) with baseline Glasgow Coma Scale (GCS), and 90-day modified Rankin Scale (mRS) in multivariable ordinal logistic regression models including age, sex, number of affected regions (odds ratio, OR, 95% confidence interval).

Results

From 651 patients (mean age 72 years, 53% male, median GCS 14), 628 CTs were analysed, 23 were excluded (not processed or contrast-enhanced). Not all cases had required data available. Student-software agreement with reference standard was κ=0.81, with diagnostic accuracy (n=314) of sensitivity 84.01%, specificity 97.88%, positive 96.67% and negative 85.92% predictive values. Using student-software results: worse GCS (n=388) was associated with intraventricular haemorrhage (OR=0.26, 0.15-0.46) and number of affected compartments (OR=0.61, 0.44-0.84); worse mRS (n=436) was associated with lobar, deep, posterior fossa, intraventricular haemorrhage, and number of affected compartments (OR range=2.22-6.92).

Conclusions

An inexperienced CT reader with AI software achieved substantial agreement and diagnostic accuracy with reference standard for brain haemorrhage location, and identified clinically relevant outcomes.

Trial Registration Number

Not applicable

Background And Aims

Drowsiness and sedentary behaviour are common accompaniments to a wide range of systemic and neurological illnesses. However, it is unclear whether such information could be used within an Early Warning System. We developed a method for quantifying behaviours, and measured whether this behavioural measure improved prediction of clinical deterioration in stroke in-patients.

Methods

Participants wore 4 inertial movement unit watches on wrists and ankles, allowing them to engage in everyday activities. Watch data was paired with annotated videos to train a deep learning model, enabling classifications of five behaviours (lying flat, sitting up in bed, sitting upright, standing or walking). Subjects were either instructed to perform a series of behaviours, or left to act as they wished. Cross-validation F1 scores for instructed sessions were: 0.95 (SE ±0.01; n=18); and for free-living: 0.88 – 0.74 (SE ± 0.04 -0.11; n=41). Subsequently we applied our system to stroke inpatients during daytime hours and assessed whether behavioral quantification was an independent predictor of clinical deterioration.

Results

Of 61 recruited subjects, 22 (36%) deteriorated over the following 5 days. The strongest clinical predictors of deterioration (stroke severity, baseline mobility, brain hemorrhage) together enabled a predictive accuracy of 72% (AUROCC: 0.64); while adding daytime behavioral metrics – particularly time spent in bed - increased this to 78% (AUROCC: 0.79).

Conclusions

In summary, we show that behavioral changes are common prior to clinical deterioration in stroke patients; often before other more classical signs of deterioration; and can be quantified using smartwatches.

Trial Registration Number

Not applicable.

Background And Aims

Magnetic Particle Imaging (MPI) enables real-time tracking of magnetic nanoparticles and offers a wide range of potential medical applications. MPI is particularly well suited for imaging blood flow and thus vessels and organs. As MPI devices can be utilized to be low power mobile devices the technology is suited for clinical usage where immobile imaging modalities like MRI or CT cannot be used, e.g. the intensive care unit.

Methods

With optimized pick-up coils MPI is suitable to detect the 3D flow dynamic of a tracer into the brain without gating techniques at very high frame rates (46 frames/s). We investigate the impact of the new pick up coil in a mouse model in case of an ischemic stroke and hemorrhage. Furthermore, we present a prototype of a clinical scale MPI stroke detection system.

Results

The increased sensitivity of the coil not only reduces the detection limit compared to body coils, but it also allows to improve the spatial resolution, so that even small vessels and anatomical structures can be detected in a mouse model. While ischemic stroke is already well visible in the native MPI images, digital subtraction of the images is needed for the detection of hemorrhage. Furthermore, by using multi-contrast reconstruction techniques, it is even possible to distinguish liquid and coagulated blood.

Conclusions

MPI provides high resolution, high sensitivity and fast perfusion imaging. The clinical scale has proven to be feasible while the technology is on the edge to enter clinical in-vivo imaging.

Trial Registration Number

Not applicable

Background And Aims

Models to predict stroke risk typically look at the whole population. We examine differences in feature importance by age-group.

Methods

We train machine learning models (logistic regression, random forest) using the risk factor datasets from Framingham-Cohort and Framingham-Offspring data from the NHLBI for four age-groups (under-50, 50-59, 60-69, over-70) with six features (sex, systolic blood pressure, diabetes, BMI, blood-pressure treatment, smoking). We sample two controls for each stroke and use 20% of the data for testing and 80% for training.

Results

Across the age-groups, logistic regression achieved AUC, F1, and Spiegelhalter's p-values of {0.75,0.67,0.51}, {0.81, 0.36, 0.78}, {0.73, 0.36,0.20}, {0.68, 0.22, 0.53}, and random forest {0.69,0.25,0.67}, {0.61,0.46,0.93}, {0.69, 0.45,0.05}, {0.57, 0.40, 0.16}. The F1 scores are low and AUC and p-values are high suggesting good calibration, but a suboptimal threshold.

In all models, systolic blood pressure is a top feature. The other top features are: Logistic regression under-50 {smoking, diabetes, sex}, 50-59 {blood-pressure treatment, smoking, diabetes}, 60-69 {sex, smoking, blood-pressure treatment}, 70+ {BMI, diabetes, blood-pressure treatment}. Random forest under-50 {BMI, smoking, sex}, 50-59 {BMI, blood-pressure treatment, smoking}, 60-69 {BMI, sex, smoking}, 70+ {BMI, diabetes, sex}.

Conclusions

Our results show there are differences in feature importance by age that should be considered in predicting stroke risk.

This work was supported by: PRECISE4Q Predictive Modelling in Stroke project funded from the EU’s Horizon 2020 research and innovation programme grant agreement No. 777107; ADAPT Research Centre, funded under the SFI Research Centres Programme (Grant 13/ RC/2106) co-funded under the European Regional Development Funds.