Figure 7 .

Comparisons of the classification performance using different thalamic partitions in identifying DOC from HC and MCS from VS. The classification accuracy of separating DOC and HC was improved by 6% when using the features of thalamic subfields, compared with the whole thalamus based classification. For the classification of MCS and VS, the accuracies were around 60% using the features of the whole thalamus or the features of thalamic subfields derived from a group-level parcellation (i.e., FOD template and thalamic connectivity atlas; Behrens et al. 2003). In contrast, the classification accuracy was improved to 80% when using the individualized thalamic partitions. The error bars represent the standard deviation of classification accuracies across the 3-fold cross-validation process of each classification task. The distribution of subjects represented in the feature space based on the top 3 principal components shows a clear boundary between DOC patients and controls and a larger separation between MCS and VS groups when using individualized thalamic partitions relative to the group-level partitions.