Context. The retina, a sophisticated neural network for the visual system, plays a crucial role in the transduction of photonic signals from the external environment into a cascade of electrical impulses transmitted to the brain. This complex organ, situated at the posterior segment of the ocular structure, is endowed with the capability to execute advanced tasks such as differential motion detection (Gollisch-Meister, 2010) and motion anticipation (Menz et al., 2020) with superior efficiency compared to contemporary computational devices. Despite these capabilities, comprehensive understanding of retinal functionality remains an enigma. A crucial challenge lies in elucidating the dynamics of the retinal network, particularly the response of retinal ganglion cells (RGCs)– the pivotal cells responsible for transmitting the retina’s output to the brain – to diverse visual stimuli. Current methodologies employ stimuli with elementary spatiotemporal structures, enabling extensive sampling across various frequencies and contrasts within the temporal domain. However, these approaches are constrained by limited spatial resolution (Baden et al., 2016). The utilization of “natural” images as stimuli introduces additional complexity, due to their extensive informational content, thereby making it difficult to identify specific stimulus features cellular responses. Moreover, prevailing data processing techniques for cellular classification predominantly rely on model-based algorithms, with a substantial oversight of the intricate retinal network structure. The prevailing paradigm predominantly emphasizes localized cellular information processing, with minimal consideration for dynamic network processes such as activity waves generated by mobile stimuli (Souihel & Cessac, 2019, Cessac, 2022). This scenario underscores the imperative for innovative approaches and methodologies that holistically encompass both the spatial and temporal dimensions, thereby fostering a more comprehensive and nuanced understanding of retinal functionality and dynamics.
Objectives. We want to propose new classes of stimuli and stimulation protocols to better understand the functional structure of the retina from the measurement of RGCs spatiotemporal responses. This is a transdisciplinary challenge. On the experimental side, one needs to develop efficient simulation-acquisition protocols, using a feedback stimulation-control loop to precisely stimulate retinal ganglion cells in a determined region of space, with a time-dependent stimulus adapting online to the measured retinal responses. On the modeling side, the goal is to infer and simulate a complex virtual network that mimics the retina behavior, composed of multiple hidden layers. For this, we have only information about the input, the light simulation, and the output, the measurement of retinal ganglion cells. We want to develop models, grounded on biophysics, with a minimal set of parameters, and a good mathematical control of what is going on [Kartsaki et al, 2023], in contrast to AI “black boxes” approaches. AI can be part of the process, but it will not be the main tool. Instead, we will concentrate on developing mathematical methods and efficient algorithms to infer and simulate, in real-time, a virtual network from which we will propose stimuli that will be tested online on real retinas. The virtual network and the stimuli will be adapted according to the observed retinal responses.
Partners.
Adrián Palacios, Center Interdisciplinary for Neuroscience of Valparaiso (CINV), Universidad de
Valparaiso (Chile), Full professor, permanent position, http://cinv.uv.cl/apalacios/ and Instituto
de Sistemas Complejos de Valparaíso, http://sistemascomplejos.cl
Maria Jose Escobar, Advanced Center for Electrical and Electronic Engineering (AC3E)
Universidad Federico Santa Maria, Chile
