Fig 1.
An anatomically and functionally based model describing the in- and output stream of the right primary visual area (V1): Besides the primary auditory pathway (red arrows), auditory information coming from the cochlear thread at the level of the tectum, inferior and superior colliculus (IC, SC) (an audiovisual interface), and the secondary visual pathway, pulvinar (Pv), into V1 (blue arrows). Further, sensory areas, A1 and V1, modulate via the thalamus (Tha) the supplementary motor area (pre-SMA) in order to optimize temporal processing and predictive coding (cerebello-thalamic-pre-SMA loop). Investigating this network, four areas (green) were used for the DCM analyses. BG = basal ganglia, Cb = cerebellum, LGN = lateral geniculate nucleus, MGN = medial geniculate nucleus.
Table 1.
Clinical and behavioral data of the vision-impaired and normally sighted subjects.
Fig 2.
Volumes of interest used for the model space definition.
Location of the domain in which blind subjects showed their individual peak coordinates (yellow) within cytoarchitectonic masks of (A) pulvinar (Pv), (B) supplementary motor area (pre-SMA), (C) primary visual area (V1), and (D) primary auditory area (A1). Note the Pv peaks did neither overlap with MGN (light blue) nor lateral geniculate nucleus (LGN) (dark blue). All selected structures overlaid on a T1 template.
Fig 3.
Model comparison was performed on family-level inferences: Models were subgrouped into four families specified by (Family 1) coupling between V1/A1 and SMA, (Family 2) coupling between A1 and SMA, (Family 3) coupling between V1 and SMA, and (Family 4) those with/without coupling between A1 and SMA, but without any connection to/from Pv/V1. Within the families, models consisted of either (i) bidirectional, (ii) absent, or (iii) unidirectional connectivity regarding the Pv/V1 coupling to/from the remaining areas. Furthermore, unidirectional models were configured hierarchically in the way that Pv projected forward to A1/V1 which again projected forward to SMA. Backward projections were only defined with respect to bidirectional models, not in an unidirectional way. A priori, driving input on Pv/A1, bidirectional A1-V1, and absent Pv-pre-SMA connectivity was assumed.
Fig 4.
Bayesian model selection (BMS).
(a) Exceedance probabilities and posterior expectations (in parentheses) resulting from the BMS procedure. The analysis did not reveal a clear winning family. But, with high confidence (total exceedance probability, p = 0.83), pre-SMA received either input from V1 (Fam. 3) or from A1, while Pv/V1 are not connected with the remaining areas (Fam. 4). (b) Posterior model probabilities for all subjects, assigned to the families (see also Table 2), indicating that blind subjects preferably chose Fam. 3, whereas Fam. 4 was much more likely for sighted individuals. More specifically, blind subjects showed high probability for the model m11, whereas sighted individuals primarily chose m13 or m27.
Table 2.
Posterior model probabilities for subjects and models resulting from the family-level inferences (BMS).
Bold numbers indicate the most likely model (in Occam’s window) for each subject.
Fig 5.
Bayesian model averaging (BMA).
Individual DCM mean parameters of (a) driving input and (b) connection strength, tested for significance (parameter ≠ 0, one sample t test) separately for each subgroup (blind and sighted). Consistently for both subgroups and all speech conditions (bw8, bw16, fw8, fw16), driving input on A1 and Pv was highly significant. Significant values are represented by asterisks: p < 0.05 (*), p < 0.01 (**), p < 0.001 (***). Lower panels show Bonferroni Holm corrected data of blind (c) and sighted (d) individuals applied to the anatomically/functionally based network hypotheses (gray). Driving input is exemplified with the forward ultra-fast speech condition (fw16).
Table 3.
BMA results of the family-level BMS procedure (across blind and sighted).
Parameter estimates (mean, standard error) of the driving input and intrinsic connectivity were calculated for each subgroup, blind and sighted, separately. Italic numbers indicate significance (p < 0.05), bold italic numbers indicate significance under Bonferroni Holm correction (connectivity: p < 0.005, driving input: p < 0.006).
Fig 6.
Correlation between DCM parameter and performance of ultra-fast speech comprehension.
Connection strength of V1-pre-SMA (unfilled squares) and driving input on Pv (filled diamonds) plotted against individual behavioral performance of ultra-fast speech comprehension of the blind subgroup. Regression lines, correlation coefficients (Spearman Rho), and significance level were given.