Cardiac arrest & resuscitation
DeepCRI: real-time EEG-based prognostication after cardiac arrest
Authors' conclusion
By embedding DeepCRI into routine ICU EEG infrastructure, we demonstrate the technical feasibility and clinical promise of continuous, real-time AI-driven prognostication for comatose patients after cardiac arrest.
Study context
- Population
- Comatose patients after cardiac arrest with EEG in the first 36 hdevelopment dataset
- Participants
- 522 patients
- Comparison
- Internal validation vs External validation
- Internal validationn = 219
- External validationn = 167
- Outcome
- Accuracy of DeepCRI lock-in classification for good and poor neurological outcome
- Follow-up
- First 36 h after arrest
Methodological detailsmulticentre development dataset, internal and external validation, deep learning
- Multicentre development dataset
- Internal and external validation
- Deep learning
Performance
As reported in the abstract
DeepCRI lock-in classification
| Test or model | Value | Sensitivity, good outcome | Specificity, good outcome | Sensitivity, poor outcome | Specificity, poor outcome | Gray zone |
|---|---|---|---|---|---|---|
| Internal validation | 81.7% | 94.7% (95% CI 90.0-97.6%) | 81.9% (95% CI 73.5-88.1%) | 49.5% (95% CI 40.2-58.9%) | 100.0% (95% CI 96.7-100.0%) | 40/219 (18.3%) |
| External validation | 59.9% | 67.2% (95% CI 54.7-77.7%) | 82.1% (95% CI 73.7-88.2%) | 36.8% (95% CI 28.2-46.3%) | 98.4% (95% CI 91.3-99.7%) |
Value
Internal validation81.7%
External validation59.9%
Sensitivity, good outcome
Internal validation94.7% (95% CI 90.0-97.6%)
External validation67.2% (95% CI 54.7-77.7%)
Specificity, good outcome
Internal validation81.9% (95% CI 73.5-88.1%)
External validation82.1% (95% CI 73.7-88.2%)
Sensitivity, poor outcome
Internal validation49.5% (95% CI 40.2-58.9%)
External validation36.8% (95% CI 28.2-46.3%)
Specificity, poor outcome
Internal validation100.0% (95% CI 96.7-100.0%)
External validation98.4% (95% CI 91.3-99.7%)
Gray zone
Internal validation40/219 (18.3%)
External validation— not stated in the abstract
Further findings
False poor-outcome prediction (post hoc analysis)
- Residual EMG artifacts contributed to the false poor-outcome prediction
- Post hoc
Original publication
DeepCRI: real-time EEG-based prognostication after cardiac arrest. Resuscitation. Published 16 April 2026. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →