Deep Learning Versus Neurologists: Functional Outcome Prediction in LVO Stroke Patients Undergoing Mechanical...

Area of Science:

• Neuroscience
• Medical Imaging
• Artificial Intelligence

Background:

• Functional recovery after large vessel occlusion stroke is variable and difficult to predict.
• Current treatments for stroke have limitations in predicting patient outcomes.

Purpose of the Study:

• To improve the estimation of functional outcomes in large vessel occlusion stroke patients.
• To evaluate interpretable deep learning models for stroke outcome prediction using clinical and MRI data.

Main Methods:

• Collected data from 222 patients with middle cerebral artery M1 segment occlusion undergoing mechanical thrombectomy.
• Utilized 5-fold cross-validation to assess deep learning models predicting modified Rankin Scale at 3 months.
• Compared model performance against 5 experienced stroke neurologists using clinical and imaging data.

Main Results:

• The deep learning model using clinical and diffusion-weighted imaging showed the highest binary prediction performance (AUC 0.766).
• Models integrating imaging data significantly outperformed neurologists (72% accuracy vs. 64% accuracy).
• Perfusion-weighted imaging did not enhance outcome prediction accuracy.

Conclusions:

• Interpretable deep learning models can significantly improve early prediction of functional outcomes in stroke patients.
• AI-assisted prediction may enhance clinical decision-making for large vessel occlusion stroke.
• Deep learning models show promise in supporting neurologists for better stroke outcome prediction.