Welcome to the WSC 2022 Interactive Program
The congress will officially run on Singapore Standard Time (SGT/UTC+8)
To convert the congress times to your local time Click here
*Please note that all sessions in halls Summit 1, Summit 2 & Hall 406 will be live streamed in addition to the onsite presentation
ASK THE SPEAKER
Sessions in Halls 406, Summit 1 and Summit 2 have a Q&A component, through the congress App called “Ask the Speaker”
MACHINE LEARNING APPROACHES TO PREDICT POSTSTROKE DEPRESSION DURING THE SUBACUTE STAGE OF STROKE
Abstract
Background and Aims
Poststroke depression (PSD) is a major psychiatric disorder that develops after stroke; however, whether PSD treatment improves cognitive and functional impairments is not clearly understood. In this study, we aimed to use various ML algorithms to predict the occurrence and prognosis of PSD in stroke patients based on their cognitive and functional status and evaluated whether ML algorithms are superior to statistical methods.
Methods
We reviewed data from 31 subjects with PSD and 34 age-matched controls without PSD; all subjects underwent neurological, cognitive and functional assessments, including the National Institutes of Health Stroke Scale (NIHSS), Korean version of the Mini-Mental Status Examination (K-MMSE), computerized neurocognitive test (CNT), Korean version of the Modified Barthel Index (K-MBI), and functional independence measure (FIM) at admission to the rehabilitation unit in the subacute stage following stroke and 4 weeks after initial assessments. Machine learning methods, such as support vector machine, k-nearest neighbors, random forest, and voting ensemble models, and statistical analysis using logistic regression were performed.
Results
PSD was successfully predicted using a support vector machine linear algorithm [area under curve (AUC) = 0.711, accuracy = 0.619]. PSD prognoses could be predicted using a support vector machine with a radial basis function kernel function (AUC = 0.830, accuracy = 0.771). The statistical method did not have a better AUC than that of machine learning algorithms.
Conclusions
We concluded that the occurrence and prognosis of PSD in stroke patients can be predicted effectively based on patients’ cognitive and functional statuses using machine learning algorithms.
ASSESSMENT OF CEREBRAL AUTOREGULATION IN REAL TIME
Abstract
Background and Aims
The use of cross-spectral analysis of spontaneous slow oscillations of cerebral and systemic hemodynamics with the determination of phase shift (PS) reflects the state of cerebral autoregulation (CA) reliably. However, this assessment of CA is carried out retrospectively after the studies performed. At the same time, CA-oriented therapy is becoming increasingly important with obtaining data about state of CA in real time, especially in intensive care.Aim: to develop a software and hardware complex for assessment the state of CA using cross-spectral analysis in real time.
Methods
The analysis of blood flow velocity and systemic blood pressure signals obtained using Multi Dop X and CNAP was performed within a frame sliding along the signals. For the signals, received in the frame, coherent components, belonging to a given frequency range (50-150 mHz), were isolated. The PS was determined.
Results
During the examination of healthy volunteers, continuous recording of indicators was carried out for 30 minutes. The PS was in the range of 0.9˗1.4 rad. Hypercapnic and hypocapnic loads led to a significant decrease and increase of PS by 25 and 43%, respectively. During the examination of patients with atherosclerotic carotid stenosis, a decrease in PS (0.1-0.5 rad) was observed on the pathology side and less pronounced responses to capnic effects (by 10 and 20% respectively).
Conclusions
The results of the conducted studies confirm the possibility of using the software and hardware complex for continuous monitoring of CA for the purpose of diagnosis and timely correction in conditions including intensive care
CLINICAL APPLICATION OF THE REAL-TIME ARTIFICIAL INTELLIGENCE IN THE DETECTION OF HEMORRHAGIC STROKE
Abstract
Background and Aims
The study of acute stroke is increasingly utilizing image analysis powered by artificial intelligence (AI). Its use to detect and quantify bleeding suspect hyperdensities in non-contrast-enhanced head CT (NCCT) images may aid clinical decision-making and expedite stroke care
Methods
Recently commercially available a novel uAI-based algorithm was used to assess the NCCTs of 84 patients with suspected acute stroke for the presence or absence of acute intracranial hemorrhages (ICH). Three neuroradiology residents evaluated the CT scan report. A professional neuroradiologist established the final report. For the detection of ICH and intraparenchymal hemorrhage (IPH), the specificity, sensitivity, and area under the curve were calculated
Results
There were 41 cases of ICH that were positive and 43 cases that were negative. The uAI has an accuracy of 96.7 percent, a sensitivity of 94.6 percent, and a specificity of 98.2 percent. ICH was further classified into the following subtypes: intraparenchymal, intraventricular, epidural/subdural, and subarachnoid, with 93.8, 96.4, 96.3, and 91.2 percent true positive rates, respectively. The true positive rates for ICH were 75.4, 100, and 100 percent, respectively, by volume [small (5 mL), medium (15–25 mL), and big (>25 mL). There were 25 cases of positive LVO (large vessel occlusion) and 59 cases of negative LVO. The LVO tool achieved 96.5 percent accuracy, 98.4 percent sensitivity, and 98.5 percent specificity.
Conclusions
In this dataset, the AI-based approach accurately determined the presence or absence of acute ICHs and quantified IPH volumes. The software's superior performance in early time periods is most likely explained by the subtle changes
PERFORMANCE OF A MULTIMODAL AI-ENABLED STROKE DIAGNOSIS TOOL IN POSTERIOR AND NON-POSTERIOR STROKE PATIENTS
Abstract
Background and Aims
Stroke triage is a common emergency department challenge, especially for posterior stroke syndromes. We previously described a multimodal deep learning framework, DeepStroke, to assess patterns of facial incoordination and speech in patients with possible stroke.
DeepStroke produces a binary classification (stroke/TIA vs. non-stroke) and achieves better performance than both a triage team and ER doctors overall among all stroke types. The primary aim for this study was to determine non-inferiority for sensitivity of detection PCA versus non-PCA strokes.
Methods
We enrolled 82 consecutive patients with stroke symptoms. Patients with obvious strokes who do not require triage are excluded. We determined the gold standard diagnosis to be the vascular neurolgist's discharge diagnosis. Strokes were then categorized as having any ischemia in the PCA territory (PCA) or not (non-PCA) based on DWI MRI. We analyzed the specificty, sensitivity, and accuracy of the tool to determine if DeepStroke had a tendency to under-diagnose strokes arising from the posterior circulation. We chose a non-inferiority margin of 5% for each task.
Results
For strokes with PCA ischemia, the sensitivity, specificity, and accuracy of diagnosis by DeepStroke was 75%, 58%, 66%. For non-PCA strokes the sensitivity, specificity, and accuracy of diagnosis was 77%, 58%, 68%. These results fell within the non-inferiority bounds.
Conclusions
In this consecutive cohort of ischemic stroke patients, DeepStroke perfomed similarly in both PCA and non-PCA stroke patients and remains superior to the triage nurse decision in both stroke types. Application of this tool may provide a valuable means for detecting stroke at triage.