A 3D Convolutional Neural Network to Model Retinal Ganglion Cell’s Responses to Light Patterns in Mice

Abstract

Deep Learning offers flexible powerful tools that have advanced our understanding of the neural coding of neurosensory systems. In this work, a 3D Convolutional Neural Network (3D CNN) is used to mimic the behavior of a population of mice retinal ganglion cells in response to different light patterns. For this purpose, we projected homogeneous RGB flashes and checkerboards stimuli with variable luminances and wavelength spectrum to mimic a more naturalistic stimuli environment onto the mouse retina. We also used white moving bars in order to localize the spatial position of the recorded cells. Then recorded spikes were smoothed with a Gaussian kernel and used as the output target when training a 3D CNN in a supervised way. To find a suitable model, two hyperparameter search stages were performed. In the first stage, a trial and error process allowed us to obtain a system that is able to fit the neurons firing rates. In the second stage, a systematic procedure was used to compare several gradient-based optimizers, loss functions and the model’s convolutional layers number. We found that a three layered 3D CNN was able to predict the ganglion cells firing rates with high correlations and low prediction error, as measured with Mean Squared Error and Dynamic Time Warping in test sets. These models were either competitive or outperformed other models used already in neuroscience, as Feed Forward Neural Networks and Linear-Nonlinear models. This methodology allowed us to capture the temporal dynamic response patterns in a robust way, even for neurons with high trial-to-trial variable spontaneous firing rates, when providing the peristimulus time histogram as an output to our model.

Visual percepts evoked with an intracortical 96-channel microelectrode array inserted in human occipital cortex

“…The visual encoder system has been described elsewhere (12,13,(55)(56)(57)(58)(59). In brief it consisted of a video camera attached to an eyeglass frame for image acquisition using head scanning, and custom hardware/software which performs a real-time analysis of the light patterns that are received by the light sensors in the camera and a multichannel spatio-temporal filtering of the visual information to extract and enhance the most relevant features of the scene.…”

Machine Learning Method for Functional Assessment of Retinal Models

“…Parameter values are selected, so as to produce different perspectives of the same image and, at the same time, preserve the object for detection inside the frame of the image. Data augmentation is implemented using imaug library 1 .…”

“…This model was chosen, as it can effectively predict retinal responses to natural images and, being trained with natural images, it can model a wide range of retina's biological properties. To train the retinal model, we used an image dataset consisting of 4890 grayscale natural images of size 50x50 pixels and the recorded retinal responses (retinal responses were recorded at Prof. E. Fernandez lab) [1]. Each frame, corresponding to 10 ms of visual stimulus, was projected onto the retina of a mouse for a total of 50 ms.…”

“…The increasing knowledge of visual systems along with technological advances give novel results in prevention, limitation or even treatment of visual impairment [1], [2]. However, retinal degenerative diseases, such as age-related macular degeneration (AMD) and retinitis pigmentosa cannot be effectively treated with surgery or medication [3].…”

“…Towards this direction, the development of retinal models aims to simulate the biological neural processing in the retina, by interpreting images to retinal responses. Nowadays, advanced retinal models incorporate computer vision techniques [5], [6] and, more recently, state-of-the-art models use Convolutional Neural Networks (CNNs) [1], [2].…”