Clinical Applications of Machine Learning in Stroke Care

Abstract

Stroke is one of the leading causes of death and disability worldwide. In recent years, machine learning (ML) methods have been increasingly applied to stroke care. This  research work  reviewed studies using ML approach in stroke care to provide an overview of this field. The advances of ML techniques have made it possible to automatically examine carotid plaques (for stroke risk stratification), detect stroke lesions on imaging, identify possible treatment complications in patients, facilitate brain-computer-interface (BCI)-aided rehabilitation, and predict stroke prognosis. The performances of certain machine learning applications are non-inferior to clinicians in areas such as measuring carotid intima-media thickness and detecting early damage of ischemic stroke on CT imaging. In addition, ML applications in clinical outcome prediction have similar or better performances than the conventional method logistic regression. However, there are still challenges in areas such as automated lesion segmentation, BCI-aided rehabilitation and long-term stroke prognosis prediction. Newly developed ML methods such as deep learning may be promising to overcome the challenges. Further research is needed to verify and optimize these ML applications, and large-sample studies and proper validation are warranted to make these ML methods more accurate, generalizable and reliable. As the need for precision medicine in stroke grows and as the technology of machine learning advances, it is anticipated that the potential of machine learning applications will be released to improve computer-aided stroke prevention and transform conventional stroke medicine into data-driven personalized stroke management, which will reduce morbidity and mortality rates, enhance stroke care and set patients free from stroke-caused disability.