Machine learning-based classification of cortical response to visual stimuli recorded with an ECoG array in mice: a case study

Optic nerve has been proven to be a valid target for neuroprosthetic implants to restore vision in blind individuals. The optimization of stimulation parameters is crucial for the correct functioning of such implants. Closed-loop stimulation of the optic nerve exploiting the cortical activation pattern as feedback may be a valid optimization strategy. Animal experiments are needed to decode the visual cortex activity in response to simple visual stimuli with a view to testing, in vivo, closed-loop stimulation of the optic nerve. The design of such experiments must take into account the possibility of extending the study to blind human subjects. In this paper, we propose to use machine learning to classify the cortical activation of a mouse, recorded with an electrocorticography (ECoG) array, in response to visual stimuli. The mouse was anesthetized, and 10 different visual stimuli were displayed on a screen positioned 20 cm away from its right eye, while the ECoG array recorded the response of the mouse's left primary visual cortex. We tested 3 classifiers (Random Forest, support vector machine, and multilayer perceptron) and obtained above-chance classification accuracy (47% with 10 classes), showing interesting preliminary results of ECoG and machine learning in terms of visual cortex decoding.