Deep learning model integrating radiologic and clinical data to predict mortality after ischemic stroke.

Area of Science:

• Neurology
• Radiology
• Artificial Intelligence

Background:

• Existing prognostic indexes for ischemic stroke mortality often lack crucial radiologic information.
• Accurate mortality prediction is vital for timely intervention and patient management.

Purpose of the Study:

• To develop and validate a deep learning model for predicting ischemic stroke mortality.
• To integrate brain diffusion-weighted imaging (DWI), apparent diffusion coefficient (ADC) data, and clinical factors into a novel prediction model.

Main Methods:

• A deep learning algorithm (DLP_DWI) was initially trained using DWI and ADC data from ischemic stroke patients.
• Clinical factors were incorporated to create an integrated model (DLP_INTG).
• Model performance was evaluated using time-dependent AUC (TD AUC) and C-index for one-year mortality prediction on internal and external datasets.

Main Results:

• The DLP_DWI model showed moderate predictive performance (TD AUC: 0.643 internal, 0.785 external).
• The integrated DLP_INTG model significantly outperformed existing scores in both internal (TD AUC: 0.859) and external (TD AUC: 0.876) datasets.
• Both models demonstrated strong discrimination for identifying patients at high risk of one-year mortality.

Conclusions:

• Deep learning models integrating neuroimaging (DWI, ADC) and clinical data can effectively predict one-year mortality in ischemic stroke patients.
• The developed DLP_INTG model offers a more accurate and comprehensive prognostic tool compared to traditional methods.
• This approach holds promise for improving risk stratification and guiding treatment decisions in acute ischemic stroke care.