Stroke is a leading cause of morbidity and mortality worldwide, with acute ischemic and hemorrhagic strokes representing the two main categories. Accurate prediction of stroke outcomes is crucial for tailoring treatment strategies and providing patients with realistic expectations for their recovery. Traditionally, prognosis relied heavily on clinical judgment and experience, which could vary among healthcare providers. The arrival of artificial intelligence (AI) and machine learning (ML) has brought about a new age of precision medicine, offering exciting prospects for objective and precise predictions.
AI, defined as the ability of computers to simulate intelligent behavior, has made remarkable strides in various fields, including healthcare (Krittanawong et al., 2017; Garcia-Vidal et al., 2019; Schwalbe and Wahl, 2020; Bonkhoff and Grefkes, 2022). Machine learning techniques, such as logistic regression (LR), random forests (RF), support vector machines (SVM), and deep learning (DL), have demonstrated significant potential in dealing with unstructured data, such as medical images (Deo, 2015; Esteva et al., 2017). These algorithms can combine data characteristics (features) with flexible decision boundaries in non-linear ways, making them well-suited for complex medical data analysis.
While AI has already demonstrated its prowess in fields like radiology and dermatology (Gulshan et al., 2016; Esteva et al., 2017; Hannun et al., 2019), its application in stroke prognosis remains relatively unexplored. The prognosis of a stroke is difficult to predict accurately because it depends on many factors specific to each patient and their clinical condition (Mendelson and Prabhakaran, 2021; Toyoda et al., 2022).
A systematic review and meta-analysis conducted by Yang et al. (2023) sought to assess the predictive accuracy of AI in stroke prognosis. A study published on September 7, 2023 in the Frontiers in Neuroscience journal thoroughly analyzed literature from three esteemed databases, PubMed, Embase, and Web of Science. The researchers examined data up until February 2023, ensuring that the study’s findings are based on the most up-to-date and relevant information available. To be included, studies had to involve cohorts with complete data, predictive models based on AI, and recognized prognosis records. The study’s primary outcome measure was the area under the curve (AUC), a commonly used indicator of predictive model accuracy.
Out of 1,241 publications, the study ultimately included seven studies involving 4,379 ischemic stroke patients. Notably, the risk of bias was low across the selected studies, and there was no significant heterogeneity. The pooled AUC, under a fixed-effects model, was 0.872 with a 95% confidence interval (CI) of 0.862–0.881. This result suggests that AI models show promising accuracy in predicting ischemic stroke outcomes.
The AI models used in the studies encompassed various algorithms, including DL, LR, RF, SVM, and Xgboost. The DL subgroup demonstrated an AUC of 0.888, while LR achieved an AUC of 0.852. RF and SVM subgroups scored AUCs of 0.863 and 0.905, respectively. The Xgboost subgroup reached an AUC of 0.905 (Yang et al., 2023).
These findings highlight the potential of AI in improving stroke prognosis accuracy. The choice of algorithm also plays a role in prediction performance, with SVM and Xgboost showing superior results. Notably, these AI models leverage patient-specific and clinical data, enhancing their clinical applicability.
However, it’s essential to consider the feasibility and acceptability of AI-based prediction tools in clinical practice. While DL models may offer high predictive accuracy, their complexity and lack of interpretability may limit their adoption. Future research should prioritize improving AI algorithms for clinical applications. One potential approach is to integrate these algorithms into electronic health records, enabling real-time decision support (Yang et al., 2023).
Despite the promising results, this study has some limitations. It primarily used AUC as the sole indicator of predictive accuracy, not accounting for sensitivity, specificity, or accuracy rates. Additionally, the sample sizes in the analyzed studies were relatively small, emphasizing the need for larger datasets in future AI predictive models.
In conclusion, AI has demonstrated high accuracy in predicting ischemic stroke outcomes, making it a valuable tool for clinicians to assess patient prognosis. The choice of AI algorithm can significantly impact predictive performance, with SVM and Xgboost models standing out. As AI algorithms continue to evolve and incorporate big data, they are likely to provide even better predictive models for stroke prognosis, ultimately benefiting patients and healthcare providers alike.