JMIR Publications

ABSTRACT

Background:

Post-stroke seizures and post-stroke epilepsy are important complications that affect recovery and long-term quality of life after stroke. Predicting which patients are at higher risk remains challenging, and recent studies have explored the use of artificial intelligence to improve prediction.

Objective:

This scoping review aimed to map how artificial intelligence and related computational methods have been applied to predict post-stroke seizures and epilepsy, to describe the role of neuroimaging in these models, and to assess representation of African settings in the existing literature.

Methods:

A systematic search identified studies that used statistical, machine learning, radiomics, or deep learning approaches to predict seizure-related outcomes after stroke. Traditional statistical models and clinical risk scores were the most commonly used methods. A smaller number of studies applied machine learning techniques, while radiomics and deep learning approaches were reported in only a few studies. Neuroimaging was mainly used to predict long-term epilepsy rather than early seizures. Most models were evaluated using internal validation only, with limited external validation.

Results:

Studies from African settings were underrepresented and primarily focused on describing seizure frequency and associated factors using regression analysis. No studies were identified that developed or validated artificial intelligence-based prediction models using African datasets.

Conclusions:

Overall, the findings show that while interest in AI-based prediction is growing, advanced methods remain limited, and major gaps exist in model validation and geographic representation. These findings highlight the need for context-specific, well-validated prediction models, particularly in African healthcare settings.