Isolation forest-voting fusion-multioutput: A stroke risk classification method based on the multidimensional output...

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Public Health

Background:

• Stroke is a leading global health threat with high incidence, fatality, and recurrence rates.
• Current electronic medical record (EMR) based stroke screening faces challenges in accuracy and risk level recognition due to systematic errors and data collection issues.
• Subjectivity in evaluation indicators and potential misreporting further complicate accurate stroke risk assessment.

Purpose of the Study:

• To develop an advanced algorithm for improved stroke screening and risk prediction using EMR data.
• To address limitations in current stroke risk assessment methods by incorporating a novel computational approach.
• To provide multidimensional auxiliary decision-making information for healthcare professionals.

Main Methods:

• A novel isolation forest-voting fusion-multioutput algorithm was developed and applied to processed and normalized screening data.
• The composite feature score index was utilized to analyze the importance of various stroke risk factors.
• The algorithm identifies abnormal samples and performs classification, outputting risk factor importance, abnormal sample labels, risk level, and stroke prediction.

Main Results:

• The proposed algorithm categorizes stroke risk into five levels: zero, low, high, transient ischemic attack (TIA), and hemorrhagic stroke (HE).
• The average accuracy for stroke prediction using this model reached 79.59%.
• The model provides multidimensional outputs, including risk factor importance, abnormal sample identification, risk stratification, and stroke prediction.

Conclusions:

• The isolation forest-voting fusion-multioutput algorithm effectively improves stroke risk level identification and prediction accuracy.
• The model's ability to output multidimensional auxiliary information aids medical staff in clinical decision-making.
• This approach significantly enhances the efficiency and accuracy of stroke screening processes.