Machine Learning on Stroke Risk Prediction Systems as Complementary Technology for Neurologists: A Critical Review

Stroke is considered the second-leading cause of death all over the world. Due to the risks entangled with stroke, the need to identity its early signs and symptoms is vital in preventing it. By recognizing the modifiable and non-modifiable risk factors related to stroke, numerous systems are developed through a machine learning (ML) approach. Ergo, utilizing stroke-risk prediction systems could prevent its occurrence and minimize fatality. This literature review focuses on assessing the current models and systems available. Numerous studies stick to utilizing traditional machine learning algorithms although the machine learning pipelines of the reviewed studies are mostly in question due to the lack of performing data pre-processing techniques. On the other hand, there are several studies which develop new machine learning algorithms based on traditional algorithms to perform more specific predictions. There has also been a perceived increase in studies wherein graphical user interface (GUI) is integrated in the process of presenting the results of a prediction system and this trend helps deliver the established systems to be used in the healthcare industry. Although that seems to be the end goal, implementation of machine learning algorithms still can be improved by setting guidelines and reporting standards to increase the acceptability and face validity of the algorithms’ predictions. Factor assessment is also key in addressing issues surrounding the transparency of machine learning by showing how risk factors have contributed to the prediction. Overall, further study and continued research on its reliability and dependability may aid in the forward trajectory of modern diagnosis.