Fig. 1: Individualized prognosis for DoC using CeTh features.

a Magnetic resonance imaging (MRI) coronal view showing electrode placements in the bilateral central thalamus (CeTh) of an example patient (Patient No. 18), with an inset showing the left hemisphere electrode (L, left; R, right; A, anterior; P, posterior). b Visualization of electrophysiological signals extracted from the raw recording: neuronal spikes and multiunit activity (MUA). Examples from two patients are shown: Patient No. 4 (consciousness non-recovery, abbreviated as unCR) is characterized by a predominantly silent state, exhibiting only background activity, while Patient No. 21 (consciousness recovery, abbreviated as CR) displays intermittent bursting activity and corresponding unstable MUA activity. The normalized power spectrum (PSD) for the MUA signals of both patients is presented, with the theta band highlighted. c Electrophysiological profiles of the CeTh before and after feature selection: the left panel shows the distribution of 34 CeTh features across patients. The importance of Single-Feature (Isf) is indicated by the Fisher Score (circle size) and effect size (color intensity), illustrating the variability among individual features. The four selected features are highlighted with black borders and are further detailed in the right panel, including stability in theta band (MUAstabilityTheta, abbreviated as stab-θ), theta band power (MUApowerTheta, abbreviated as pow-θ), spiking-MUA synchronization in the gamma band (syncMUAGamma, abbreviated as sync-γ), stability in the high-gamma band (MUAstabilityHGamma, abbreviated as stab-hγ). The importance in Multi-Feature Combination (Imf) represents the stable-importance index (circle size) and feature weights (color intensity) in the CeTh metric as shown in (f). d Permutation test: the accuracy achieved with the actual dataset and the distribution of accuracies from Permuted Dataset. e Feature contribution evaluation: progressive improvement in model accuracy and F1 score as features are sequentially added (with the sequence of stab-θ, pow-θ, sync-γ, stab-hγ, see “Methods”), contrasted with baseline performance near chance level (50%) in the permuted datasets. f Decision values for all 23 patients from both the machine learning model and the CeTh metric. Values above 0 are classified as consciousness recovery (CR), while values below 0 are classified as non-recovery (unCR). The dashed line represents the model’s classification threshold. Vertical connecting lines show the difference between each patient’s decision values as determined by the model and the CeTh metric. Source data are provided as a Source Data file.