An artificial intelligence-based prognostic prediction model for hemorrhagic stroke.

Area of Science:

• Neurology
• Medical Imaging
• Artificial Intelligence

Background:

• Intracerebral hemorrhage (ICH) carries a poor prognosis.
• Existing prognostic models for ICH lack comprehensive imaging features.
• Accurate prediction of short-term outcomes is crucial for patient management.

Purpose of the Study:

• To develop a novel clinic-imaging fusion model using convolutional neural networks (CNNs).
• To predict the short-term neurofunctional prognosis of patients with ICH.
• To evaluate the model's performance against established ICH scales.

Main Methods:

• A multi-center retrospective study involving 1990 ICH patients.
• Construction of two CNN-based deep learning models for outcome prediction.
• Validation using nested 5-fold cross-validation and comparison with ICH and ICH grading scales.

Main Results:

• The clinic-imaging fusion CNN model achieved the highest AUC (0.903) in the test set.
• An imaging-based CNN model showed strong predictive performance (AUC = 0.886).
• Both CNN models significantly outperformed the ICH scale (AUC = 0.777) and ICH grading scale (AUC = 0.747).

Conclusions:

• CNN-based prognostic models using neuroimaging are superior to traditional ICH scales for predicting discharge outcomes.
• Integrating clinical data further enhanced the predictive efficiency of the CNN model.
• This approach offers a more effective tool for predicting neurological outcomes in ICH patients.