Structural testing of multi-input linear-nonlinear cascade models for cells in macaque striate cortex

Structural testing methods based on experimental white noise stimulus-response data were used to evaluate multi-input linear-nonlinear (LN) cascade models for simple and complex cells in macaque striate cortex. An LN structural test index, based on white noise stimulation, was developed and found to be suitable for classifying cells as simple vs complex. In particular, classification results based on the LN structural test index were similar to classification results based on a traditional modulation index derived from cell responses to drifting sinewave gratings. Judging from their structural test indices, complex cells deviated more strongly from LN behavior than did simple cells. Yet, even with simple cells, on average, only about 60% of the first- and second-order white noise stimulus-response relation was consistent with LN behavior. Just two of thirteen simple cells studied had an LN consistency level that exceeded 80%. Similar results were found in tests for consistency with an LNL model which includes an additional linear post-filter. We conclude that a conventional multi-input LN network model may be a useful approximation to the response behavior of some simple cells. However, even during steady state stimulus conditions, subcortical and/or cortical nonlinearities other than a static output nonlinearity play a very significant role in shaping the responses of most simple cells in the macaque striate cortex.