Machine Learning-Based Prediction for In-Hospital Mortality After Acute Intracerebral Hemorrhage Using Real-World
Koutarou Matsumoto1,2, Kazuaki Ishihara3, Katsuhiko Matsuda4
1Department of Health Care Administration and Management, Graduate School of Medical Sciences Kyushu University Fukuoka Japan.
Journal of the American Heart Association
|December 10, 2024
Summary
Machine learning (ML) models effectively predict early mortality in acute intracerebral hemorrhage (ICH) patients. Incorporating specialist expertise significantly enhances the predictive performance of these ML models for improved patient outcomes.
Area of Science:
- Neurology
- Medical Imaging
- Machine Learning
Background:
- Machine learning (ML) is increasingly used for accurate predictions across domains.
- Acute intracerebral hemorrhage (ICH) poses significant mortality risks.
- Evaluating ML for predicting early mortality in ICH patients is crucial.
Purpose of the Study:
- To assess the effectiveness of ML in predicting in-hospital mortality risk for ICH patients.
- To compare ML model performance against traditional risk scores.
- To determine the impact of specialist input on ML model accuracy.
Main Methods:
- Developed ML models using brain CT imaging and clinical data from 527 ICH patients.
- Evaluated model performance using AUC, calibration plots, and decision curve analysis.
- Compared ML models with ICH score and ICH grading scale.
Main Results:
- ML models showed good calibration and differentiated survival rates.
- ML model using clinical and specialist-assessed image data achieved the highest AUC (0.97).
- ML models using raw imaging or non-specialist clinical data also demonstrated satisfactory performance.
Conclusions:
- ML-based models are effective for predicting ICH mortality using imaging or clinical data.
- Specialist expertise significantly enhances the predictive accuracy of ML models for ICH.
- ML offers a promising tool for early mortality risk assessment in ICH.
