(Upper) Blind-to-controls generalization performance: prediction accuracies observed when the classifier was trained on the auditory trials from the blind participants and tested on either the auditory or the visual trials from the sighted controls. (Upper) Blind-to-controls generalization performance: prediction accuracies observed when the classifier was trained on the auditory trials from the blind participants and tested on either the auditory or the visual trials from the sighted controls. (Lower) Controls-to-blind generalization performance: prediction accuracies observed when the classifier was trained on either the visual or the auditory trials from the sighted controls and tested on the auditory trials from the blind subjects. The red dashed line indicates chance level (50%) and is not used for assessment of statistical significance. An asterisk (*) in a bar reflects a significant decoding accuracy compared with the empirically estimated chance level. *P < 0.05, FDR-corrected at q = 0.05. Job van den Hurk et al. PNAS doi:10.1073/pnas.1612862114 ©2017 by National Academy of Sciences