Table 4. Comparison of Fully Automated Identification of Epileptiform Discharges Between SCORE-AI and encevis on the Same 100 EEGs Used in the Multirater Test Set.
| Algorithm | Fully automated identification of EEG recordings with epileptiform abnormalities, % (95% CI) | ||||
|---|---|---|---|---|---|
| Sensitivity | Specificity | Negative predictive value | Positive predictive value | Accuracy | |
| encevisa | 96.68 (88.89-100.00) | 27.14 (17.19-37.88) | 95.03 (83.33-100.00) | 36.29 (25.93-46.99) | 48.03 (38.00-58.00) |
| SCORE-AIb | 89.94 (77.78-100.00) | 87.13 (78.79-94.29) | 95.28 (89.39-100.00) | 74.98 (60.00-88.57) | 87.97 (81.00-94.00) |
| Difference (P value) | <.001 | <.001 | .49 | <.001 | <.001 |
Abbreviations: EEG, electroencephalography; SCORE-AI, Standardized Computer-based Organized Reporting of EEG–Artificial Intelligence.
a
For encevis, the detection of one or more spikes was considered as epileptiform classification of the EEG.
b
For SCORE-AI, either epileptiform-focal and/or epileptiform-generalized was considered as an epileptiform classification of the EEG.