Deep learning-based personalised outcome prediction after acute ischaemic stroke.

Area of Science:

• Artificial Intelligence in Medicine
• Neuroimaging and Clinical Data Integration
• Cardiovascular and Cerebrovascular Event Prediction

Background:

• The ability of deep learning models to predict individual long-term risk of major adverse cerebro/cardiovascular events (MACE) after acute ischemic stroke (AIS) using clinical data and brain imaging remains unstudied.
• Accurate prediction of MACE is crucial for effective patient management post-AIS.

Purpose of the Study:

• To investigate the efficacy of deep learning models in predicting individual long-term MACE risk following AIS.
• To compare the performance of deep learning models against traditional survival models using both clinical and neuroimaging data.

Main Methods:

• A cohort of 8590 patients with AIS was analyzed for MACE (stroke, myocardial infarction, death) within 12 months.
• Deep learning models (DeepSurv, DeepSM) and traditional models (CoxPH, RSF) were evaluated using time-dependent concordance index.
• Model performance was assessed with and without the inclusion of brain imaging features.

Main Results:

• Deep learning models incorporating brain imaging significantly outperformed traditional models.
• DeepSurv and DeepSM achieved the highest concordance index (0.8496 and 0.8531) when brain images were combined with clinical factors.
• Brain imaging features were highly important, with deep learning models automatically extracting relevant information for personalized risk prediction.

Conclusions:

• Deep learning models integrating clinical data and brain imaging enhance the prediction of MACE in AIS patients.
• These models facilitate personalized outcome prediction, offering a significant advancement over traditional prognostic systems.
• The findings support the development of more accurate and tailored risk prediction tools for AIS management.