Interpretability Analysis of One-Year Mortality Prediction for Stroke Patients Based on Deep Neural Network
IEEE Journal of Biomedical and Health Informatics
|October 29, 2021
Summary
This study introduces an interpretable deep learning model for predicting stroke patient mortality risk. The model utilizes Bidirectional Long Short-Term Memory with a novel attention module, achieving high accuracy in predicting one-year mortality.
Area of Science:
- Neurology
- Artificial Intelligence
- Medical Informatics
Background:
- Clinical data collection is crucial for stroke patient prognosis.
- Accurate prediction of one-year mortality risk in stroke patients is essential for clinical decision-making.
Purpose of the Study:
- To develop an interpretable deep learning model for predicting one-year mortality risk in stroke patients.
- To reconstruct clinical features highlighting variable dissimilarity and temporality.
Main Methods:
- Utilized Bidirectional Long Short-Term Memory (Bi-LSTM) architecture.
- Incorporated a novel correlation attention module to consider variable interdependencies.
- Trained the model on a clinical dataset of 2,275 stroke patients.
Main Results:
- Achieved high performance metrics: precision of 0.9414, recall of 0.9502, and F1-score of 0.9415.
- Demonstrated the model's interpretability through visualizations aligned with clinical guidelines.
Conclusions:
- The developed interpretable deep learning model shows significant promise for accurate stroke mortality risk prediction.
- The model's ability to reconstruct and analyze clinical data features enhances prognostic capabilities.
