Prediction of hemorrhagic transformation after experimental ischemic stroke using MRI-based algorithms

Area of Science:

• Neuroimaging
• Stroke Research
• Medical Diagnostics

Background:

• Hemorrhagic transformation (HT) risk estimation is critical for acute ischemic stroke treatment decisions.
• Accurate prediction of HT is essential to guide therapeutic interventions and improve patient outcomes.

Purpose of the Study:

• To evaluate the accuracy of multiparametric MRI-based predictive algorithms for calculating HT probability post-stroke.
• To compare the performance of different predictive models, including generalized linear models and random forest algorithms.

Main Methods:

• Utilized spontaneously hypertensive rats subjected to embolic stroke and treated with tissue plasminogen activator or vehicle.
• Acquired multiparametric MRI data (T2, T2*, diffusion, perfusion, BBB permeability) at multiple time points post-stroke.
• Developed and validated generalized linear model and random forest (RF) algorithms for HT and infarct prediction using acute MRI data.

Main Results:

• A RF-based model incorporating spatial brain features demonstrated high accuracy in predicting hemorrhage (AUCs of 0.85 ± 0.14 and 0.89 ± 0.09).
• This RF model significantly outperformed perfusion- or permeability-based thresholding methods for hemorrhage prediction.
• Overlap between predicted and actual tissue outcomes was lower for hemorrhage prediction (max DSI 0.20 ± 0.06) compared to infarct prediction (max DSI 0.81 ± 0.06).

Conclusions:

• Multiparametric MRI-based predictive algorithms can effectively identify tissue at risk of HT early after stroke.
• These algorithms hold potential for improving treatment decision-making in acute ischemic stroke management.
• Further refinement of these models may enhance the precision of HT risk assessment.