Clinical Scorecard: Utilizing Deep Learning to Forecast Clinical and Tissue Outcomes of Thrombectomy in Acute Ischemic Stroke
At a Glance
Category
Detail
Condition
Acute ischemic stroke with intracranial large vessel occlusion
Key Mechanisms
Endovascular thrombectomy (EVT) to restore reperfusion; deep learning models predict tissue salvage and clinical outcomes
Target Population
Patients with acute ischemic stroke undergoing thrombectomy
Care Setting
Acute stroke care, primarily in hospital settings with multimodal CT imaging capabilities
Key Highlights
Deep learning models predict individual tissue and clinical outcomes post-thrombectomy under scenarios of successful and unsuccessful reperfusion.
Models trained on a large dataset (405 patients) using multimodal CT imaging and clinical data outperform traditional thresholding and machine learning methods.
Predictions include penumbral salvage and NIH Stroke Scale (NIHSS) score changes, offering personalized biomarkers for treatment benefit and resource allocation.
Guideline-Based Recommendations
Diagnosis
Use multimodal CT imaging for initial assessment of acute ischemic stroke with large vessel occlusion.
Incorporate patient-specific clinical characteristics alongside imaging for outcome prediction.
Management
Consider endovascular thrombectomy as a standard treatment for eligible patients with large vessel occlusion.
Utilize predictive models to estimate individual benefit from thrombectomy to guide personalized therapeutic strategies.
Monitoring & Follow-up
Assess NIH Stroke Scale (NIHSS) at discharge to evaluate clinical outcomes post-thrombectomy.
Monitor infarct growth and reperfusion success to inform prognosis.
Risks
Recognize variability in patient response to thrombectomy, with some patients experiencing progressive infarct growth despite intervention.
Be aware of disparities in access to EVT, especially in low- and middle-income regions.
Patient & Prescribing Data
405 ischemic stroke patients with intracranial large vessel occlusion undergoing thrombectomy
Deep learning models provide superior prediction accuracy for tissue and clinical outcomes compared to traditional methods, enabling tailored treatment decisions.
Clinical Best Practices
Integrate advanced predictive modeling using deep learning with routine multimodal CT imaging to improve individualized outcome forecasts.
Use NIHSS as a quantitative measure for clinical outcome assessment rather than solely relying on categorical scales like mRS.
Address geographical and infrastructural disparities by optimizing resource allocation based on predicted thrombectomy benefits.
by Marie-Sophie von Braun, Kristin Starke, Lucas Peter, Daniel Kürsten, Florian Welle, Hans Ralf Schneider, Max Wawrzyniak, Daniel P O Kaiser, Gordian Prasse, Cindy Richter, Elias Kellner, Marco Reisert, Julian Klingbeil, Anika Stockert, Karl-Titus Hoffmann, Gerik Scheuermann, Christina Gillmann, Dorothee Saur
