Tissue at Risk and Ischemic Core Estimation Using Deep Learning in Acute Stroke.

Area of Science:

• Neurology
• Medical Imaging
• Artificial Intelligence

Background:

• Acute ischemic stroke with large vessel occlusions requires predicting final infarct size.
• Reperfusion therapy outcomes are crucial for patient management.
• Accurate estimation of tissue at risk and ischemic core is essential.

Purpose of the Study:

• To demonstrate the value of deep learning for estimating tissue at risk and ischemic core.
• To compare deep learning approaches with conventional methods.
• To improve prediction of infarct evolution after reperfusion therapy.

Main Methods:

• Trained attention-gated convolutional neural networks using baseline MR imaging from 3 multicenter trials.
• Explored separate, pretraining/fine-tuning, and thresholding approaches.
• Evaluated models using Dice score coefficient and area under the curve.

Main Results:

• The pretraining approach achieved the highest median Dice score for tissue at risk (0.60) and core (0.57).
• Deep learning models significantly outperformed conventional thresholding methods.
• Pretraining demonstrated superior performance compared to separate models and thresholding.

Conclusions:

• Deep learning models with fine-tuning enhance prediction of tissue at risk and ischemic core.
• This AI-driven approach surpasses conventional thresholding in accuracy.
• Improved prediction aids in optimizing reperfusion therapy for stroke patients.