Multidimensional machine learning for early neurological deterioration prediction in acute ischemic stroke

Area of Science:

• Neurology
• Artificial Intelligence
• Biomedical Informatics

Background:

• Early neurological deterioration (END) is a critical concern in acute ischemic stroke (AIS).
• Accurate prediction of END risk is essential for timely therapeutic interventions.
• Existing prediction models may not fully capture the complexity of AIS progression.

Purpose of the Study:

• To develop and validate a multidimensional clinical feature-based machine learning model for predicting END risk in AIS patients.
• To identify key clinical and pathological indicators associated with END.
• To compare the performance of different machine learning algorithms for END prediction.

Main Methods:

• A cohort of 338 AIS patients was divided into training (n=236) and validation (n=102) sets.
• Five core predictors were identified: NIHSS score, blood glucose, infarct core volume, collateral circulation, and NLR.
• Random Forest (RF), Gradient Boosting Machine (GBM), and K-Nearest Neighbors (KNN) models were constructed and evaluated using AUC.

Main Results:

• Multivariate logistic regression identified NIHSS score, blood glucose, infarct core volume, and NLR as independent risk factors for END.
• Collateral circulation status was identified as an independent protective factor.
• The RF model achieved the highest predictive performance with AUCs of 0.779 (training) and 0.775 (validation).

Conclusions:

• The developed multidimensional machine learning model shows potential for early clinical identification of high-risk END patients.
• Timely intervention can be facilitated by accurate END risk prediction.
• The RF model demonstrates superior efficacy in predicting END in AIS patients compared to GBM and KNN.