Introduction: Stroke is a leading cause of disability worldwide, and precise identification of stroke lesions is essential for prognosis and rehabilitation planning. Machine learning has emerged as a powerful tool for automating stroke lesion segmentation and predicting post-stroke recovery, though challenges in standardization and generalizability remain. Methods: This narrative review synthesizes studies published between 2010 and 2024 on the application of machine learning in stroke lesion segmentation and recovery forecasting. A total of 23 relevant studies were reviewed, including 10 focused on lesion segmentation and 13 on recovery prediction. Results: Deep learning models such as U-Net and con- volutional neural networks(CNNs) have shown improved segmentation accuracy using the Anatomical Tracings of Lesions After Stroke (ATLAS) dataset, though dataset bias and inconsistent evaluation metrics limit comparability. Recovery prediction studies demon- strated the use of lesion characteristics alongside clinical and demographic features to forecast motor, cognitive, and functional outcomes. However, issues such as small sample sizes, lack of multi-centre validation, and heterogeneous modelling approaches were noted. Discussion: The review highlights the interdependence between segmentation and predic- tion—accurate lesion delineation informs recovery modelling, while recovery prediction may enhance the clinical relevance of segmentation. Integration of both domains is critical to advancing stroke care. Conclusion: Machine learning offers promising advances in stroke lesion analysis and recovery forecasting. Future efforts should focus on standardized datasets, multimodal data integration, and robust validation to translate research into clinical practice and support precision rehabilitation.
Simi et al. (Mon,) studied this question.