gms | German Medical Science

Introduction: One of the most challenging issues for epiretinal prosthetics is how to optimize multi-focal electric stimuli (MFES) so that they induce visual percepts with desired spatial characteristics. This is an intractable issue for experiments with human subjects due to the multi-dimensional stimulus space, time limitations, and subject fatigue and frustration. To address this problem, we developed an in-vivo animal (cat) model in which the spatial information transmitted by a MFES paradigm is assessed by decoding cortical (area V1) population responses. Our approach is based on the premise that MFES paradigms that maximize the spatial information transmitted to cat V1 are likely to be the most efficacious at producing visual percepts that correspond to the stimulation patterns.

Methods: An 8x10 intracortical recording array inserted in V1 records multi-site local field potentials (ms-LFPs). The ms-LFP tuning for retinotopic location, orientation, and spatial frequency is determined using flashed (40 msec) visual stimuli. A Multi-Taper, Space-Frequency Singular Value Decomposition method is applied to the ms-LFP signal to identify dominant patterns of spatial coherence within different local frequency bands. This analysis also allows meaningful comparisons of visual vs. electric ms-LFP responses, which may differ in absolute latencies and/or durations. The spatial coherence patterns identified in a subset of the stimulation trials are used to train a Support Vector Machine (SVM) that learns to associate cortical response dynamics with inducing stimuli. Subsequently, the SVM is used to decode stimuli from the response dynamics induced in a non-overlapping subset of trials. This analysis computes the joint probability distribution between delivered and decoded stimuli, which determines the amount of information transmitted to V1. MFES are delivered via custom-made, thin-film 32-electrode arrays implanted epiretinally at the region providing input to the monitored V1 sites. Stimulating electrodes (flat-contact, 60 µm diameter platinum disks, separated by 65 µm), are arranged in a 6x6 grid (minus 4 corners), and the return electrode is on the animal’s ear. Current pulses are charge-balanced, cathodic-first (typically: 0.25 msec x 4-12 µAmps). MFES are realized by delivering identical pulses to groups of electrodes along different orientations (0°, 30°, 60°, 90°, 120°, 150°). In this study, the inter-electrode pulse latency (IEPL) was varied to examine the dependence between temporal structure of MFES and orientation information transmitted to V1.

Results: (1) Spatial position, orientation, and spatial frequency of visual stimuli were all decoded with high resolution from the V1 ms-LFP. Thus, the V1 ms-LFP captures a large amount of the spatial information transmitted by the retina. (2) The orientation of MFES with IEPL=0 (simultaneous pulse injection) was not decoded accurately from the V1 ms-LFP signal. Conversely, the orientation of MFES with IEPLs of 3 and 5 msec was decoded accurately. Therefore, temporal dispersion of MFES is crucial for increasing the spatial information transmitted by an implant.

Conclusions: Objective assessments of the efficacy of different paradigms of retinal prosthetic stimulation are achievable using our cortical ms-LFP-decoding model. Further characterizations are feasible based on the correspondence between visual and electric ms-LFP responses.

Support: National Science Foundation grant to N.P. Cottaris (CBET-0756098) and Ligon Research Fund.

This lecture is available as video recording (Attachment 1 [Attach. 1]).