Fig. 4.

Correlations are not driven by lesions. a) A potential confound in the analyses of patient data is that the correlations between observed and predicted maps might be driven by lesions. This concern is motivated by the fact that lesions are the same in each subject's observed and predicted maps (shown here for one example subject with the lesion mask in black); and because little to no neural activity was observed within the lesion masks. To check that the results were not driven by lesions, we repeated the main analyses above while excluding the union of all lesion masks (see Supplementary Fig. 4). b) The lesions did not strongly affect the results, as shown by the similarity matrix (between observed and predicted language maps). For example, a heavy diagonal was again found in both controls (upper-left) and patients (lower-right quadrant). c) The repeated t-test analysis was very similar as before, with 90/103 individuals identified better than baseline (cf. 93/103). Of the three subjects previously identified but not identified here, two were control subjects. Consequently, removing the lesion masks resulted in a decrease in accuracy of only one patient.