Send to

Choose Destination
Eur J Neurosci. 2019 Jul 8. doi: 10.1111/ejn.14505. [Epub ahead of print]

Individualized Quantification of the Benefit from Reperfusion Therapy using Stroke Predictive Models.

Author information

Neurobiology Research Unit and Center for Integrated Molecular Brain Imaging, Copenhagen, The Neuroscience CentreRigshospitalet, Denmark.
Department of Biostatistics, University of Copenhagen, Øster. Farimagsgade 5, PB 2099, DK 1014, Copenhagen K.
Department of Stroke Medicine and Department of Neuroradiology, Université Lyon 1, CREATIS, CNRS, UMR 5220-INSERM U1044, INSA-Lyon, Hospices Civils de Lyon, Lyon, France.
Center of Functionally Integrative Neuroscience, Århus University, Århus, Denmark.
Klinik und Poliklinik für Neurologie, Kopf- und Neurozentrum, Universitätsklinikum Hamburg-Eppendorf, Hamburg, Germany.
Department of Radiology (IDI), Girona Biomedical Research Institute (IDIBGI), Hospital Universitari de Girona Dr Josep Trueta, Av de Francia s/n, 17007, Girona, Spain.
Service de Biostatistique, Hospices Civils de Lyon, Lyon, France.
Equipe Biostatistique Santé CNRS UMR 5558, Villeurbanne, France, Université Lyon I, Lyon, France.
Department of Clinical Neurosciences, University of Cambridge, Cambridge, UK; Dept of Neurology, Hôpital Sainte-Anne, Paris Descartes University, INSERM U894, Paris, France.



Recent imaging developments have shown the potential of voxel-based models in assessing infarct growth after stroke. Many models have been proposed but their relevance in predicting the benefit of a reperfusion therapy remains unclear. We searched for a predictive model whose volumetric predictions would identify stroke patients who are to benefit from tissue Plasminogen Activator (t-PA)-induced reperfusion.


Forty-five cases were used to study retrospectively stroke progression from admission to end of follow-up. Predictive approaches based on various statistical models, predictive variables, and spatial filtering methods were compared. The optimal approach was chosen according to the area under the precision-recall curve (AUPRC). The final lesion volume was then predicted assuming that the patient would or would not reperfuse. Patients with an acute lesion of ≤50 mL and a predicted reduction in presence of reperfusion >6 mL, and >25% of the acute lesion were classified as responders.


The optimal model was a logistic regression using the voxel distance to the acute lesion, the volume of the acute lesion, and Gaussian-filtered MRI contrast parameters as predictive variables. The predictions gave a median AUPRC of 0.655, a median AUC of 0.976, and a median volumetric error of 8.29mL. Nineteen patients matched the responder profile. A non-significant trend of improved reduction in NIHSS score (-42.8%, p=0.09) and in lesion volume (-78.1%, p=0.21) following reperfusion was observed for responder patients.


Despite limited volumetric accuracy, predictive stroke models can be used to quantify the benefit of reperfusion therapies. This article is protected by copyright. All rights reserved.


magnetic resonance imaging; predictive modeling; reperfusion; stroke


Supplemental Content

Full text links

Icon for Wiley
Loading ...
Support Center