Fig. 4.

Canonical microcircuit (CMC) model for DCM and connections modulated by stimulus noise.

The left panel illustrates the forward model of how cell populations interact both within and between cortical areas, and how their changing conductances generate the MEG signal. DCM inverts this model and estimates the connectivity parameters describing the strength of forward and backward connections (light blue arrows) and intrinsic or self-inhibitory connections (circled in blue), which control the excitability or gain of the superficial pyramidal cell population. The sensory input, if present in a given area, enters the spiny stellate population, here illustrated with a green arrow. The CMC model improves upon past forward models that only contain three populations – i.e. those that group superficial and deep pyramidal cells together – because superficial and deep layers are the sources of forward and backward connections (i.e. of prediction errors and predictions, in predictive coding) respectively (Bastos et al., 2012). Reproduced from a personal communication from Dr Harriet Brown.

The right panel shows the DCM structure, with the (forward and intrinsic) connections that could be modulated by noisy target motion highlighted in blue. Sensory inputs enter all three sensory sources. Please note that lateral connections between identical areas were included at all levels (including between central V1 and bilateral cuneus; i.e., peripheral V1), but are not shown for clarity. In DCM, extrinsic forward, backward and lateral connections are excitatory, while self-connections are inhibitory. Excitatory connections end in arrows, and inhibitory connections in balls. Cun — cuneus, i.e. part of V1 representing the peripheral visual field.