Individualized quantification of the benefit from reperfusion therapy using stroke predictive models.

Area of Science:

• Neurology
• Medical Imaging
• Biostatistics

Background:

• Voxel-based models show promise in assessing stroke-related infarct growth.
• The predictive value of existing models for reperfusion therapy benefit remains unclear.

Purpose of the Study:

• To identify a predictive model for quantifying stroke patients who benefit from tissue plasminogen activator (t-PA)-induced reperfusion.
• To assess the relevance of volumetric predictions in guiding reperfusion therapy decisions.

Main Methods:

• Retrospective analysis of 45 stroke cases, comparing predictive approaches using statistical models and spatial filtering.
• Optimal model selected based on area under the precision-recall curve (AUPRC).
• Responder profile defined by acute lesion volume, predicted reperfusion reduction, and percentage of acute lesion reduction.

Main Results:

• An optimal logistic regression model utilized voxel distance, acute lesion volume, and Gaussian-filtered MRI contrast parameters.
• The model achieved a median AUPRC of 0.655 and AUC of 0.976, with a median volumetric error of 8.29 ml.
• Nineteen patients met the responder profile, showing a trend towards improved NIHSS score and lesion volume reduction post-reperfusion.

Conclusions:

• Predictive stroke models, despite volumetric limitations, can effectively quantify the potential benefits of reperfusion therapies.
• These models aid in identifying stroke patients most likely to respond to t-PA treatment.