OEDL: an optimized ensemble deep learning method for the prediction of acute ischemic stroke prognoses using union...

Area of Science:

• Medical Imaging and Artificial Intelligence
• Neurology and Stroke Medicine
• Biomedical Data Science

Background:

• Accurate early prognosis assessment is crucial for timely therapeutic decisions in stroke patients.
• Current methods may benefit from integrated approaches combining diverse data types.
• Deep learning offers potential for sophisticated predictive modeling in stroke care.

Purpose of the Study:

• To develop and evaluate an integrated deep learning model for acute ischemic stroke (AIS) prognosis prediction.
• To assess the value of combining clinical and radiomics features for enhanced prediction accuracy.
• To explore the application of data combination, method integration, and algorithm parallelization in stroke prognosis.

Main Methods:

• Extracted and selected clinical (17 features) and radiomics (19 features) data from 441 stroke patients.
• Developed an Optimized Ensemble of Deep Learning (OEDL) method integrating multiple deep learning techniques.
• Employed a metaheuristic algorithm for efficient parameter optimization and utilized hybrid sampling (SMOTEENN) for data balancing.

Main Results:

• The OEDL method with combined features and hybrid sampling achieved superior classification performance.
• Achieved 97.89% Macro-AUC, 95.74% ACC, 94.75% Macro-R, 94.03% Macro-P, and 94.35% Macro-F1.
• Combined features significantly outperformed models using only clinical or radiomics data.

Conclusions:

• The OEDL approach demonstrates significant potential for improving stroke prognosis prediction.
• Integrated modeling using combined clinical and radiomics data offers superior predictive value over single-feature models.
• This method provides valuable clinical decision support for optimizing early intervention and personalized stroke treatment.