Acute cardiac care & resuscitation
Interpretable machine learning model for imaging-based outcome prediction after cardiac arrest
Authors' conclusion
We developed an interpretable method to identify patterns of early post-cardiac arrest brain injury on CT imaging and showed these imaging patterns are predictive of patient outcomes (i.e., survival and awakening status).
Study context
- Population
- Comatose adults after resuscitation from in- and out-of-hospital cardiac arrest with unenhanced brain computerized tomography (CT) within 24 h, single academic centre, Aug 2011–Aug 2019
- Participants
- 1,284 adults
- Outcome
- Prediction of patient outcomes (survival and awakening status), measured by AUC
Methodological detailssingle-centre, interpretable machine learning, 80%-20% random data split
- Single-centre
- Interpretable machine learning
- 80%-20% random data split
Main finding
As reported in the abstract
Prediction of survival and awakening
| Group | AUC, survival | AUC, awakening |
|---|---|---|
| Machine learning models | 0.710 ± 0.012 | 0.702 ± 0.053 |
Machine learning models
AUC, survival0.710 ± 0.012
AUC, awakening0.702 ± 0.053
Observational data: associations, not effects.
Further findings
Outcomes in cohort
- Awakened from coma
- 35%
- Survived to hospital discharge
- 34%
Clinical relevance of imaging patterns
- Expert physicians could visualise decomposed image patterns and identify clinically relevant ones in multiple brain locations
Original publication
Interpretable machine learning model for imaging-based outcome prediction after cardiac arrest. Resuscitation. Published 4 July 2023. doi:10.
Summarised from the published abstract; every number is checked against it