Table 3:
Performance of clinical, imaging and combined models predicting initial EEG state among comatose survivors of cardiac arrest.
Predicted outcome(s) | ||||
---|---|---|---|---|
Model/Inputs | Benign vs. suppressed | Benign vs. highly pathological | Suppressed vs. highly pathological | Benign vs. suppressed vs. highly pathological |
Clinical covariates | ||||
LR | 0.73 (0.02) | 0.80 (0.04) | 0.74 (0.08) | 0.74 (0.04) |
SVM | 0.69 (0.04) | 0.70 (0.05) | 0.63 (0.03) | 0.66 (0.03) |
NN | 0.73 (0.02) | 0.80 (0.05) | 0.75 (0.04) | 0.75 (0.04) |
Imaging | ||||
VGG16 | 0.69 (0.04) | 0.51 (0.03) | 0.66 (0.05) | 0.60 (0.04) |
ResNet50 | 0.64 (0.04) | 0.55 (0.06) | 0.62 (0.02) | 0.61 (0.02) |
GoogLeNet | 0.66 (0.04) | 0.51 (0.07) | 0.62 (0.04) | 0.58 (0.04) |
Combined model | NN + VGG16 | NN + ResNet50 | NN + VGG16 | NN + ResNet50 |
Performance | 0.73 (0.05) | 0.61 (0.05) | 0.67 (0.06) | 0.66 (0.03) |
Data are presented as area under the curve with standard deviations.