Section: Application Domains

Human vision understanding through joint experimental and modeling studies, for normal and distrophic retinas

Cells characterization from their spike response

A prior step toward understanding how the retina extracts the information from a visual scene is the characterization of retinal ganglion cells receptive fields. The receptive field allows to classify retinal ganglion cells in sub-types such as direction sensitive cells. Each of these type extracts a local and definite piece of information from the visual scene, transmitted to the visual cortex. Hence receptive fields are somewhat the fundamental bricks of vision.

Current techniques of receptive fields estimation are based on Spike-Triggered Average  [70]. However, this method heavily relies on the assumption that the static non linearity is convex (typically this is an exponential). Unfortunately, this violates a fundamental biophysical property of neurons: firing rate is bounded due to the refractory period. Additionally, this method is slow and of low precision.

We are working on more efficient techniques based on non-convex analysis, faster, more precise, and working for a non-convex (typically sigmoidal) non linearity. Additionnally we are also working on designing better stimuli for receptive fields estimations.

Understanding the role of spatio-temporal correlations in visual scene encoding

Retinal response to stimuli is related, on one hand, to spatio-temporal correlations of the stimulus  [76], and, on the other hand to the intrinsic spatio-temporal correlations of the retinal activity induced by its vertical and lateral connectivity  [81]. However, the role of spatio-temporal correlations in retinal coding is still controversial. With the current evolution of multi-electrode arrays recordings, it is possible to record from tens to thousands of neurons  [42], [51], [63], [86], studying not only the correlations between few neurons, but also the correlations present in a whole population of retinal ganglion cells  [73], [75], [77], [80]. The BIOVISION team has proposed a framework to study this correlation structure using Gibbs distributions (Sec. 3.2.4). Based upon the mathematical results presented in the papers [5] [45], we have developed algorithms to analyse and reproduce spatio-temporal correlations in neural assemblies containing up to a few hundreds of neurons [13], [69], [68].

We are now applying these methods for the analysis of retina data so as to better understand the role of spatio-temporal spike correlations in vision encoding.

Retinal waves

Retinal waves are bursts of activity occurring spontaneously in the developing retina of vertebrate species, contributing to the shaping of the visual system organization: retina circuitry shaping, retinotopy, eye segregation   [83], [52], [74], [53]. They stop a few weeks after birth. Wave activity begins in the early development, long before the retina is responsive to light. It was recently found that they can be reinitiated pharmacologically in the adult mammalian retina  [48]. This could have deep consequences on therapy for several degenerative retinal diseases. The mechanism of their generation, in imature, or adult retinas, remains however incompletely understood  [84].

We aim at proposing a dynamical model of retinal waves depending on a few canonical parameters (e.g. concentration of a pharmacological agent) controlling the arousal of retinal waves as well as their shape/intensity. We want, on one hand, to design a model sufficiently close to biophysics so that it can reproduce and predict experimental results, and, on the other hand, sufficiently general to provide a generic mechanisms of retinal waves arousal also describing their different types.

Trajectory anticipation, from retina to V1

Global motion processing is a major computational task of biological visual systems. When an object moves across the visual field, the sequence of visited positions is strongly correlated in space and time, forming a trajectory. These correlated images generate a sequence of local activation of the feedforward stream. At the present stage of knowledge, it is still unclear how the early visual system processes motion trajectories. Motion integration, anticipation and prediction would be jointly achieved through the interactions between feed-forward, lateral and feedback propagations within a common spatial reference frame, the retinotopic maps. Addressing this problem is particularly challenging, as it requires to probe these sequences of events at multiple scales (from individual cells to large networks) and multiple stages (retina, primary visual cortex (V1)).

In the context of the ANR Trajectory we are working on such an integrated approach. We aim at modelling the population responses at two key stages of visual motion encoding: the retina and V1 based on simultaneous micro- and mesoscopic recordings made by our partners Institut des Neurosciences de la Timone and Institut de la Vision, and design a simulator of retinal output feeding V1. This study is a step toward understanding mechanisms of motion coding and anticipation with strong impact on our understanding of the visual system.

Simulating and analysing retina's response to visual stimuli

We want to design a retina simulator integrating the most recent advances on retina modeling. We will propose a user-friendly simulator, using parallel (multi-threads) programming, in order to simulate rapidly a large piece of the retina. This platform is further described in the section Software.