CoChat

Visual Prostheses in the Era of Artificial Intelligence Technology

Open access
Sarbout, Ilias, Gungor, Ayse, Ounissi, Mehdi, Zaher, Samy, Ptito, Maurice, Kupers, Ron, Racoceanu, Daniel, Milea, Dan · 2025 · Eye and Brain

The DOI matches its Crossref record; 108 of the 108 references it cites were found in a registry or an index; 58 of the 88 sources its claims cite were checked for support.

Abstract

Background: Over the past few decades, technological advancements have transformed invasive visual prostheses from theoretical concepts into real-world applications. However, functional outcomes remain limited, especially in visual acuity. This review aims to summarize current developments in retinal and cortical prostheses (RCPs) and critically assess the role of artificial intelligence (AI) in advancing these systems. Purpose: To describe current RCPs and provide a systematic review on image and signal processing algorithms designed for improved clinical outcomes. Patients and Methods: We performed a systematic review of the literature related to AI subserving prosthetic vision, using mainly PubMed, but also, Elicit, a dedicated AI-based reference research assistant. A total of 455 studies were screened on PubMed, of which 23 were retained for inclusion. An additional 5 studies were identified and included through Elicit. Results: The analysis of current RCPs highlights various limitations affecting the quality of the visual flow provided by current artificial vision. Indeed, the 28 reviewed studies on AI covered two applications for RCPs including extraction of saliency in camera captured images, and consistency between electrical stimulation and perceived phosphenes. A total of 14 out of 28 studies involved the use of artificial neural networks, of which 12 included model training. Evaluation with data from a visual prosthesis was conducted in 7 studies, including 1 that was prospectively assessed with a human RCP. Validation with empirical data from human or animal data was performed in 22 out of 28 studies. Out of these, 15 were validated using simulated prosthetic vision. Finally, out of 22 studies leveraging a mathematical model for phosphenes perception, 14 used a symmetrical oversimplified modeling. Conclusion: AI algorithms show promise in optimizing prosthetic vision, particularly through enhanced image saliency extraction and stimulation strategies. However, most current studies are based on simulations. Further development and validation in real-world settings, especially through clinical testing with blind patients, are essential to assess their true effectiveness.

Overview

CoChat Summary

TL;DR

This systematic review examines retinal and cortical prostheses and AI/image-processing methods intended to improve prosthetic vision, identifying saliency extraction and stimulation–phosphene consistency as the two main applications. It finds most AI studies rely on simulations and oversimplified phosphene models, with limited validation in real prosthetic users.

Key findings

  • Of 455 PubMed records screened, 23 met inclusion criteria and 5 additional studies were identified via Elicit, yielding 28 AI studies; 14/28 used artificial neural networks, and 12 included model training.
  • Validation was mostly indirect: 22/28 studies used empirical human or animal data, 15 used simulated prosthetic vision, 7 used data from a visual prosthesis, and only 1 was prospectively assessed with a human RCP.
  • Among 22 studies using a mathematical model of phosphene perception, 14 used symmetrical oversimplified modeling; only 8 used asymmetric models, including 5 with an axon-map model.
  • AI applications split into saliency extraction from camera-captured images and optimization of consistency between electrical stimulation and perceived phosphenes; 12/28 studies included dynamic aspects of vision, while 16/28 considered only static measurements or images.

Methods

  • Systematic review of literature on AI, machine learning, or image processing for retinal and cortical prostheses, mainly using PubMed plus Elicit; 455 studies were screened, 23 retained, and 5 added through Elicit, for 28 total studies; search last updated January 2025.
  • One author (IS) extracted data on publication year, purpose, AI techniques, parameter tuning, phosphene assumptions, validation setting, human tasks, dynamic scenes, and electrode count, with study-by-study details in Table 2 and supplementary tables.
  • The paper also provides a narrative overview of current retinal prostheses (epiretinal, subretinal, suprachoroidal) and cortical prostheses, plus methodological, clinical, and technological challenges.

Limitations

  • Stated: most current AI studies are based on simulations; few are validated in real-world settings; only one study included an implanted human participant; oversimplified symmetrical phosphene models may inflate apparent performance and mask safety constraints; reviewed virtual-setting studies did not quantitatively assess safety limits, power draw, robustness stress-tests, uncertainty tracking, or fail-safes; heterogeneous tasks, metrics, datasets, and electrode counts hamper cross-study interpretation and benchmarking.
  • Evident from the text: screening and data extraction were performed by a single author (IS), which may limit reliability, and the synthesis is descriptive rather than meta-analytic.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.2147/eb.s524322

Cites

108 of 108 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Figures

Figure 1 Stimulation/Perception model for retinal and cortical prostheses. An input image (left) which is visible by a video camera is processed to generate an electrical stimulation via a microelectrode array (center – example with a retinal prosthesis). The electrical stimulation generates neural activity in the retinal ganglion cells, or in the visual cortex, which evokes spatially organized phosphenes, resulting in a visual percept by the blind patient (right).
Figure 1 Stimulation/Perception model for retinal and cortical prostheses. An input image (left) which is visible by a video camera is processed to generate an electrical stimulation via a microelectrode array (center – example with a retinal prosthesis). The electrical stimulation generates neural activity in the retinal ganglion cells, or in the visual cortex, which evokes spatially organized phosphenes, resulting in a visual percept by the blind patient (right).CC BY-NC · PubMed Central
Figure 2 Three different types of retinal prostheses have been implanted in the epiretinal, subretinal and suprachoroidal space of the human eye. All three technologies have demonstrated an ability to elicit phosphenes in late blind patients.
Figure 2 Three different types of retinal prostheses have been implanted in the epiretinal, subretinal and suprachoroidal space of the human eye. All three technologies have demonstrated an ability to elicit phosphenes in late blind patients.CC BY-NC · PubMed Central
Figure 3 Optimization pipeline for visual stimulation protocols. Artificial intelligence optimizes the transformation of a dynamic scene (left) into an electrical stimulation protocol delivered to an array of stimulators (middle), resulting in visual phosphenes (right) that emphasize the essential visual information and allows accurate phosphene mapping. This figure models the process using a 2500-electrode epiretinal implant, with a ganglion axon pathways model for visual phosphene computation.57
Figure 3 Optimization pipeline for visual stimulation protocols. Artificial intelligence optimizes the transformation of a dynamic scene (left) into an electrical stimulation protocol delivered to an array of stimulators (middle), resulting in visual phosphenes (right) that emphasize the essential visual information and allows accurate phosphene mapping. This figure models the process using a 2500-electrode epiretinal implant, with a ganglion axon pathways model for visual phosphene computation.57CC BY-NC · PubMed Central
Figure 4 Feature engineering and end-to-end approaches for saliency extraction from natural images.
Figure 4 Feature engineering and end-to-end approaches for saliency extraction from natural images.CC BY-NC · PubMed Central
Figure 5 (Left) Distribution of AI tools used for visual information processing in the reviewed studies. (Right) Cumulative study counts over time for the three approaches (NN – neural networks).
Figure 5 (Left) Distribution of AI tools used for visual information processing in the reviewed studies. (Right) Cumulative study counts over time for the three approaches (NN – neural networks).CC BY-NC · PubMed Central

Verification

Claims the paper makes

39 Valid · 16 Not valid · 14 Not checked · 114 Own results and interpretations

83 more claims · 13 with their cited sources checked

Create a free account to see all 183 claims and to verify the 1 reference still open.

See all claims — free account
  • It may state two separate claims as one. 54
  • Probably not a substantial claim. 55

Create a free account to see these claims.

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

4 of 4

Citations

Cites

Show all 108 references

Research works printed but not linked: 5

  1. Tassicker, 1956, Br J Physiol Opt, 13, 102
  2. Barry, 2023, Invest Ophthalmol Vis Sci, 64, 2842
  3. Huo, 2023, Invest Ophthalmol Vis Sci, 64, 5518
  4. Diaz, 2023, Invest Ophthalmol Vis Sci, 64, 5519
  5. Sahel, 2013, Invest Ophthalmol Vis Sci, 54, 1389

Metadata

Authors
Sarbout, I., Gungor, A., Ounissi, M., Zaher, S., Ptito, M., Kupers, R., Racoceanu, D., & Milea, D.
Affiliations
  • Centre National de la Recherche Scientifique (FR)
  • Hôpital Rothschild (FR)
  • Montreal Neurological Institute and Hospital (CA)
  • UCLouvain (BE)
Venue
Eye and Brain
Published
2025-08
Volume
Volume 17
Issue
6
Pages
95-113
ISSN
1179-2744 (ISSN-L)
OpenAlex
W4413808600
Cited by
3
References
101
Citations by year
2026: 3
FWCI
0.36
Type
journal article
Language
en
Open access
Locations
  • Eye and Brain (publishedVersion, cc-by-nc)
  • Europe PMC (PubMed Central) (publishedVersion, cc-by-nc)
  • PubMed (publishedVersion)
  • PubMed Central (publishedVersion, cc-by-nc)
  • Research at the University of Copenhagen (University of Copenhagen) (publishedVersion, cc-by-nc)
  • HAL (submittedVersion, other-oa)
  • Dove Medical Press (Taylor and Francis Group) (submittedVersion, cc-by-nc)
Topics
  • Life Sciences › Neuroscience › Cellular and Molecular Neuroscience › Neuroscience and Neural Engineering
  • Life Sciences › Neuroscience › Cognitive Neuroscience › EEG and Brain-Computer Interfaces
  • Life Sciences › Neuroscience › Cognitive Neuroscience › Tactile and Sensory Interactions
Keywords
retinal prosthesis, artificial intelligence, artificial neural networks, phosphenes, electrical stimulation, visual acuity, simulated prosthetic vision, blind patients

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

The claim verdicts this report counts are experimental: their accuracy is not yet measured.

A language model drew the claims from the paper on 2026-10-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-10 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Cited sources are checked with a language model’s help (gpt-6-sol).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.