Causative Classification of Ischemic Stroke by the Machine Learning Algorithm Random Forests.

Area of Science:

• Neurology
• Medical Informatics
• Machine Learning

Background:

• Identifying the etiology of acute ischemic stroke is crucial for prognosis, recurrence risk assessment, and secondary prevention.
• Different causes of stroke require tailored management strategies.
• Current methods for stroke etiology determination can be challenging.

Purpose of the Study:

• To develop and validate a machine learning (ML) model for accurately identifying the etiology of acute ischemic stroke.
• To compare the performance of various ML models in classifying stroke subtypes.

Main Methods:

• Retrospective analysis of patient data from the CASE-II study (NCT04487340) with etiologies defined by the Trial of ORG 10172 in Acute Stroke Treatment (TOAST).
• Training and evaluation of six ML models: Random Forests (RF), Logistic Regression (LR), Extreme Gradient Boosting (XGBoost), K-Nearest Neighbor (KNN), Ada Boosting, and Gradient Boosting Machine (GBM).
• Models were trained on data from October 2016 to April 2020 (phase one) and tested on data from June 2020 to December 2020 (phase two). Performance metrics included Area Under the Curve (AUC), precision, recall, accuracy, and F1 score.

Main Results:

• The Random Forests (RF) model demonstrated superior performance among the evaluated ML models.
• In the test set, the RF model achieved AUC values of 0.981 for cardioembolism (CE), 0.919 for large-artery atherosclerosis (LAA), and 0.918 for small-artery occlusion (SAO).
• Key predictors for stroke etiology included atrial fibrillation and the degree of intracranial artery stenosis.

Conclusions:

• The developed RF model serves as a valuable diagnostic tool for neurologists.
• This ML approach can assist in the accurate categorization of stroke etiologies.
• Improved etiological classification can lead to more effective stroke management and patient outcomes.