Estimating receptive fields of simple and complex cells in early visual cortex: A convolutional neural network model...

Area of Science:

• Neuroscience
• Computational Neuroscience
• Visual Neuroscience

Background:

• Primary visual cortex neurons exhibit selectivity for stimulus features like orientation and spatial frequency.
• Modeling simple cells is feasible with linear-nonlinear models, but complex cells require more intricate models due to phase invariance and subunit summation.
• Cortical neuron response variability complicates the estimation of model parameters.

Purpose of the Study:

• To develop a unified convolutional neural network (CNN) method for estimating receptive field models of both simple and complex visual cortex cells.
• To characterize the simple-to-complex nature of receptive fields using a single model parameter.
• To assess the predictive performance of the developed model for cell responses to natural images and gratings.

Main Methods:

• A CNN model incorporating a spatiotemporal filter, a parameterized rectifier unit (PReLU), and a Gaussian receptive field envelope was developed.
• A single model parameter (PReLU) was used to differentiate between simple and complex cell characteristics.
• The model was trained and validated using responses of visual cortex neurons to natural images and grating stimuli.

Main Results:

• The CNN method successfully estimated receptive field models for both simple and complex cells.
• A continuum of PReLU parameter values was observed across neurons, indicating a gradual transition from simple to complex cell properties.
• Complex-like cells showed lower response reliability, but noise ceiling analysis revealed comparable predictive performance to simple cells.

Conclusions:

• The developed CNN provides a unified framework for modeling simple and complex visual cortex cells, capturing their continuous nature.
• The findings suggest that the simple/complex dichotomy is not absolute but exists on a spectrum.
• The study validates the utility of CNNs in understanding neuronal receptive fields and their response characteristics.