Table 1.

Quantitative comparisons between Fig. 1 and possible organisational models. Comparisons were performed by Procrustes rotation. The statistical significance (probability) of the associated variance-explained (R²) statistics was assessed by repeating the Procrustes rotation after randomly permuting the area labels of the numerical models on each of 100,000 iterations while noting the fraction of times that the variance-explained statistic exceeded or was equal to the variance explained by the un-permuted organisational model. The ‘nearest-neighbour’, ‘nearest-neighbour or next-door-but-one’, ‘interhemispheric’ and the ‘combined nearest-neighbour or next-door-but-one and interhemispheric’ models were constructed by creating artificial connectivity matrices and submitting these to the same MDS procedure that was used to derive the structure shown in Fig. 1. The ‘nearest-neighbour’ and ‘nearest-neighbour or next-door-but-one’ connectivity matrices were constructed as described in Young (1992). To construct the ‘interhemispheric’ connectivity matrix, we scored all connections between homologous areas in the opposing hemispheres as ‘1’ and all other connections as ‘0’. The ‘combined nearest-neighbour or next-door-but-one and interhemispheric’ connectivity matrix was defined as the sum of the ‘nearest-neighbour or next-door-but-one’ and ‘interhemispheric’ matrices. The hierarchical model was constructed as a one-dimensional vector of shortest path-lengths through the ‘nearest-neighbour’ matrix between each area and ipsilateral V1 (as determined by Dijkstra's algorithm). The two and three streams models were constructed as one-dimensional vectors of values representing each area's stream category based on the colour-coding in Fig. 1. Note that V1, V2 and V3 were excluded from these comparisons since these areas were deemed not to belong to any particular visual processing stream; they were excluded from all comparisons reported in this table to allow for comparison across models. Although all of the modelled organisational principles appear to be reflected in Fig. 1 to some extend, the topological organisation of the human visual connectome is most parsimoniously and fully explained by the ‘combined hierarchical and three streams’ model.

Organisational Model	R2	Probability
Nearest-neighbour	.14	p = .01
Nearest-neighbour or next-door-but-one	.13	p = .02
Interhemispheric	.08	p = .05
Combined nearest-neighbour or next-door-but-one and interhemispheric	.67	p < 10−5
Hierarchical	.35	p < 10−5
Three streams	.47	p < 10−5
Combined hierarchical and three streams	.80	p < 10−5
Combined hierarchical and two streams:
ventral + lateral vs dorsal	.25	p < 10−4
dorsal + lateral vs ventral	.68	p < 10−5
ventral + dorsal vs lateral	.66	p < 10−5