Fig 4. Model performance on volunteer test cohort (n = 3).

(A) Casewise AUCs for each of the three selected models with median values shown in black. Median AUCs were 0.986, 0.988, and 0.991 for SDB, PCA, and LDA+HMM2 respectively. (B) Casewise accuracy for each of the three selected models with median values shown in black. Median accuracies were 0.910, 0.925, and 0.941 for SDB, PCA, and LDA+HMM2 respectively. (C) Model performance on a representative subject, showing the multitaper spectrogram and model-predicted probability of consciousness (P(y)) and ground-truth conscious state as recorded in [27].