The Use of Deep Learning to Predict Stroke Patient Mortality.

Area of Science:

• Neurology
• Medical Informatics
• Artificial Intelligence

Background:

• Stroke incidence is rising in aging populations, posing significant economic challenges.
• Early detection and treatment are crucial for improving stroke prognosis and patient outcomes.
• Medical service utilization and health behavior data offer a practical alternative to complex imaging for stroke detection.

Purpose of the Study:

• To develop and evaluate a deep neural network (DNN) model for stroke detection.
• To utilize readily available medical service use and health behavior data for stroke prediction.
• To compare the performance of the proposed DNN model against other machine learning methods.

Main Methods:

• A deep neural network (DNN) model was developed to analyze stroke risk.
• Principal Component Analysis (PCA) with quantile scaling was employed for feature extraction from medical records.
• The scaled PCA/DNN approach was benchmarked against five other machine learning algorithms.

Main Results:

• The study identified 15,099 patients with stroke using the developed model.
• The scaled PCA/DNN method achieved an Area Under the Curve (AUC) of 83.48%.
• This performance indicates a high degree of accuracy in stroke prediction.

Conclusions:

• The proposed scaled PCA/DNN model demonstrates significant potential for stroke detection.
• The model's reliance on accessible data makes it suitable for widespread clinical application.
• This method can serve as a valuable tool for both patients and clinicians in stroke prescreening.