Predicting functional outcome in patients with acute brainstem infarction using deep neuroimaging features

Area of Science:

• Neurology
• Artificial Intelligence
• Medical Imaging

Background:

• Acute brainstem infarctions cause significant functional impairments.
• Predicting functional outcomes is crucial for patient management.

Purpose of the Study:

• To predict functional outcomes in acute brainstem infarction patients.
• To utilize deep neuroimaging features extracted by convolutional neural networks (CNNs).

Main Methods:

• A nationwide multicenter study of 1482 patients with acute brainstem infarction.
• CNNs extracted deep neuroimaging features from diffusion-weighted imaging.
• Deep learning models were trained to predict 3-month functional outcomes (modified Rankin Scale score ≥ 3).

Main Results:

• A CNN-based model using 14 deep neuroimaging features achieved an AUC of 0.975.
• This model significantly outperformed a model using clinical, laboratory, and conventional imaging features (AUC 0.772).
• Deep neuroimaging features correlated with age, stroke severity, infarct volume, and inflammation.

Conclusions:

• Deep learning models automatically extract objective neuroimaging features from routine radiological data.
• These models accurately predict functional outcomes in brainstem infarction patients at 3 months.