Prediction of Major Stroke (NIHSS > 5) Based on MR Perfusion Imaging Using a 3D ResNet.

Area of Science:

• Neurology
• Radiology
• Artificial Intelligence in Medicine

Background:

• Clinical decision-making in ischemic stroke management, particularly after interventions like mechanical thrombectomy or in perioperative cases, benefits from accurate symptom prediction.
• Assessing stroke severity, defined by a National Institutes of Health Stroke Scale (NIHSS) score greater than 5, is crucial for patient outcomes.

Purpose of the Study:

• To develop and validate a deep learning model capable of predicting major stroke (NIHSS > 5) based on MR-perfusion imaging.
• To assess the model's performance in identifying patients with significant stroke deficits.

Main Methods:

• A retrospective analysis of 982 patients from a prospectively collected stroke registry (2015-2021) was conducted.
• An 18-layer 3D-ResNet model was trained using MR-perfusion Tmax-maps from multiple scanners to predict NIHSS > 5.
• The model's predictions were evaluated against reported NIHSS scores and their association with 90-day modified Rankin Scale (mRS) > 2 using logistic regression.

Main Results:

• The model achieved a Receiver Operating Characteristic area under the curve (ROC-AUC) of 0.87 and a Precision-Recall AUC (PR-AUC) of 0.85 on the internal test set.
• Accuracy was 78% with a weighted F1-score of 0.78 for predicting major stroke.
• Both reported and predicted NIHSS > 5 were associated with predicting a 90-day mRS > 2, indicating clinical relevance.

Conclusions:

• The developed deep learning model demonstrates feasibility and strong performance in predicting major stroke (NIHSS > 5) from MR-perfusion Tmax-maps.
• This AI-driven approach shows promise for enhancing clinical decision support in acute stroke care.