Explainable Machine Learning Predictions of Perceptual Sensitivity for Retinal Prostheses
Open accessPreprint — published as 10.1088/1741-2552/ad310f
Abstract
To provide appropriate levels of stimulation, retinal prostheses must be calibrated to an individual’s perceptual thresholds (‘system fitting’), despite thresholds varying drastically across subjects, across electrodes within a subject, and over time. Although previous work has identified electrode-retina distance and impedance as key factors affecting thresholds, an accurate predictive model is still lacking. To address these challenges, we 1) fitted machine learning (ML) models to a large longitudinal dataset with the goal of predicting individual electrode thresholds and deactivation as a function of stimulus, electrode, and clinical parameters (‘predictors’) and 2) leveraged explainable artificial intelligence (XAI) to reveal which of these predictors were most important. Our models accounted for up to 77% of the perceptual threshold response variance and enabled predictions of whether an electrode was deactivated in a given trial with F1 and AUC scores of up to 0.740 and 0.913, respectively. Deactivation and threshold models identified novel predictors of perceptual sensitivity, including subject age, time since blindness onset, and electrode-fovea distance. Our results demonstrate that routinely collected clinical measures and a single session of system fitting might be sufficient to inform an XAI-based threshold prediction strategy, which may transform clinical practice in predicting visual outcomes.
The claims and statistics of this paper haven’t been checked yet.
Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.
Integrity signals
Registry
Cites
Statistics
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
One study used machine learning to predict phosphene deactivation and threshold sensitivities in retrospectively collected data of Argus II users.
“Also, one study used machine learning to predict phosphene deactivation and threshold sensitivities in retrospectively collected data of Argus II users.95”
From Visual Prostheses in the Era of Artificial Intelligence Technology (2025) · cites this paper as [95]
The Methods (Dataset) describe a retrospective analysis of Argus II patient data, and the Discussion says the models predicted perceptual thresholds and electrode deactivation. The source does not describe predicting phosphene deactivation; it is the electrodes that were deactivated. (LLM, conf=0.85)
Experimental — accuracy not yet measured
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
Cites
XGBoost: A Scalable Tree Boosting System
Chen, T., & Guestrin, C. (2016) · Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining · doi:10.1145/2939672.2939785 · Cited by 54,569 · cited 1× · see citing works
“based on ensembles of decision trees [17] defined on multiple subsets of dataset features, allowing them to capture highly nonlinear patterns in the underlying data.”
SMOTE synthetic minority over-sampling technique
Chawla, N. V., Bowyer, K. W., Hall, L. O., & Kegelmeyer, W. P. (2002) · Journal of Artificial Intelligence Research · doi:10.1613/jair.953 · Cited by 32,865 · cited 3× · see citing works
“In each LOSO iteration, the synthetic minority oversampling technique (SMOTE) [21] was applied to the training dataset to balance the number of samples from each class.”
InfluentialA Unified Approach to Interpreting Model Predictions
Lundberg, S., & Lee, S.-I. (2017) · arXiv · doi:10.48550/arxiv.1705.07874 · Cited by 23,707 · see citing works
Peeking Inside the Black-Box: A Survey on Explainable Artificial Intelligence (XAI)
Adadi, A., & Berrada, M. (2018) · IEEE Access · doi:10.1109/access.2018.2870052 · Cited by 6,294 · see citing works
Topography of ganglion cells in human retina
Curcio, C. A., & Allen, K. A. (1990) · The Journal of Comparative Neurology · doi:10.1002/cne.903000103 · Cited by 1,849 · cited 2× · see citing works
“RGCDensity Retinal ganglion cell (RGC) density [13] RGC/deg2”
da Cruz, L., Coley, B. F., Dorn, J., Merlini, F., Filley, E., Christopher, P., Chen, F. K., Wuyyuru, V., Sahel, J., Stanga, P., et al. (2013) · British Journal of Ophthalmology · doi:10.1136/bjophthalmol-2012-301525 · Cited by 359 · cited 1× · see citing works
“, age at blindness diagnosis, age at implant surgery) from previous studies [3], [11], [12].”
The Argus® II Retinal Prosthesis System
Luo, Y. H.-L., & da Cruz, L. (2016) · Progress in Retinal and Eye Research · doi:10.1016/j.preteyeres.2015.09.003 · Cited by 351 · see citing works
Photovoltaic Restoration of Central Vision in Atrophic Age-Related Macular Degeneration
Palanker, D., Le Mer, Y., Mohand-Said, S., Muqit, M., & Sahel, J. A. (2020) · Ophthalmology · doi:10.1016/j.ophtha.2020.02.024 · Cited by 227 · cited 1× · see citing works
“With Argus II consisting of 60 electrodes, this procedure is time-consuming at best, but will quickly become infeasible as new devices are being developed that feature hundreds or even thousands of electrodes [6], [7].”
Factors affecting perceptual thresholds in epiretinal prostheses
de Balthasar, C., Patel, S., Roy, A., Freda, R., Greenwald, S., Horsager, A., Mahadevappa, M., Yanai, D., McMahon, M. J., Humayun, M. S., et al. (2008) · Investigative Opthalmology & Visual Science · doi:10.1167/iovs.07-0696 · Cited by 204 · cited 10× · see citing works
“However, these studies were either focused on a small number of subjects [2], [4] or were limited to predicting only the mean threshold across electrodes from a small number of factors [3].”
InfluentialDorn, J. D., Ahuja, A. K., Caspi, A., da Cruz, L., Dagnelie, G., Sahel, J.-A., Greenberg, R. J., McMahon, M. J., & Argus II Study Group, F. T. (2013) · JAMA Ophthalmology · doi:10.1001/2013.jamaophthalmol.221 · Cited by 177 · cited 1× · see citing works
“, age at blindness diagnosis, age at implant surgery) from previous studies [3], [11], [12].”
Data outlier detection using the Chebyshev theorem
Amidan, B., Ferryman, T., & Cooley, S. (2005) · IEEE Aerospace Conference · doi:10.1109/aero.2005.1559688 · Cited by 154 · cited 1× · see citing works
“To account for such sources of noise in threshold prediction, we discarded outlier samples using an automated, statistical method based on Chebyshev’s data distribution tail bounds [19].”
A model of ganglion axon pathways accounts for percepts elicited by retinal implants
Beyeler, M., Nanduri, D., Weiland, J. D., Rokem, A., Boynton, G. M., & Fine, I. (2019) · Scientific Reports · doi:10.1038/s41598-019-45416-4 · Cited by 143 · cited 3× · see citing works
“The horizontal raphe (A, white line) was approximated by fitting a parabola to the main vascular arcade and finding the tangent to the parabola inflexion point (adapted under CC-BY from [15]).”
Self-citationFactors Affecting Perceptual Threshold in Argus II Retinal Prosthesis Subjects
Ahuja, A. K., Yeoh, J., Dorn, J. D., Caspi, A., Wuyyuru, V., McMahon, M. J., Humayun, M. S., Greenberg, R. J., daCruz, L., & Argus II Study Group (2013) · Translational Vision Science & Technology · doi:10.1167/tvst.2.4.1 · Cited by 123 · cited 6× · see citing works
“However, these studies were either focused on a small number of subjects [2], [4] or were limited to predicting only the mean threshold across electrodes from a small number of factors [3].”
InfluentialMehta, P., Petersen, C. A., Wen, J. C., Banitt, M. R., Chen, P. P., Bojikian, K. D., Egan, C. A., Lee, S., Bałazińska, M., Lee, A., et al. (2021) · American Journal of Ophthalmology · doi:10.1016/j.ajo.2021.04.021 · Cited by 119 · see citing works
Self-citationChenais, N. A. L., Airaghi Leccardi, M. J. I., & Ghezzi, D. (2021) · Communications Materials · doi:10.1038/s43246-021-00133-2 · Cited by 110 · cited 1× · see citing works
“With Argus II consisting of 60 electrodes, this procedure is time-consuming at best, but will quickly become infeasible as new devices are being developed that feature hundreds or even thousands of electrodes [6], [7].”
Mehta, P., Petersen, C. A., Wen, J. C., et al. (2020) · bioRxiv · doi:10.1101/2020.02.26.967208 · Cited by 108 · cited 1×
“plainable artificial intelligence (XAI) relies on ML approaches that can explain why a certain prediction was made while maintaining high accuracy [9], [10].”
Self-citationFactors Affecting Perceptual Thresholds in a Suprachoroidal Retinal Prosthesis
Shivdasani, M. N., Sinclair, N. C., Dimitrov, P. N., Varsamidis, M., Ayton, L. N., Luu, C. D., Perera, T., McDermott, H. J., & Blamey, P. J. (2014) · Investigative Opthalmology & Visual Science · doi:10.1167/iovs.14-14396 · Cited by 78 · cited 9× · see citing works
“However, these studies were either focused on a small number of subjects [2], [4] or were limited to predicting only the mean threshold across electrodes from a small number of factors [3].”
InfluentialTen-Year Follow-up of a Blind Patient Chronically Implanted with Epiretinal Prosthesis Argus I
Lan, Y., Falabella, P., Christopher, P., Wuyyuru, V., Dorn, J. D., Schor, P., Greenberg, R. J., Weiland, J. D., & Humayun, M. S. (2015) · Ophthalmology · doi:10.1016/j.ophtha.2015.08.008 · Cited by 67 · cited 4× · see citing works
“However, perceptual thresholds have been shown to vary drastically not just across subjects, but also across electrodes within a subject as well as over time [2]–[5].”
Influentialpulse2percept: A Python-based simulation framework for bionic vision
Beyeler, M., Boynton, G., Fine, I., & Rokem, A. (2017) · Proceedings of the 16th Python in Science Conference · doi:10.25080/shinma-7f4c6e7-00c · Cited by 56 · see citing works
Self-citationData-driven models in human neuroscience and neuroengineering
Brunton, B. W., & Beyeler, M. (2019) · Current Opinion in Neurobiology · doi:10.1016/j.conb.2019.06.008 · Cited by 42 · see citing works
Self-citationA Computational Model of Phosphene Appearance for Epiretinal ProsthesesPreprint
Granley, J., & Beyeler, M. (2021) · bioRxiv · doi:10.1101/2021.06.07.447426 · Cited by 33 · cited 1× · see citing works
“as a proxy for disease progression, which may be important for predicting visual outcomes with patient-specific computational models of prosthetic vision [15], [22], [23].”
Self-citationThe argus ii retinal prosthesis system
Bloch, E., & da Cruz, L. (2020) · Prosthesis · doi:10.5772/intechopen.84947 · Cited by 31 · cited 1× · see citing works
“) [1], this process is part of system fitting, where perceptual thresholds are used to populate subject-specific lookup tables that determine how the grayscale values of an image recorded by the external camera are translated into electrical stimuli.”
A computational model of phosphene appearance for epiretinal prostheses
Granley, J., & Beyeler, M. (2021) · 2021 43rd Annual International Conference of the IEEE Engineering in Medicine & Biology Society (EMBC) · doi:10.1109/embc46164.2021.9629663 · Cited by 27 · see citing works
Self-citationpulse2percept: A Python-based simulation framework for bionic visionPreprint
Beyeler, M., Boynton, G. M., Fine, I., & Rokem, A. (2017) · bioRxiv (Cold Spring Harbor Laboratory) · doi:10.1101/148015 · Cited by 27 · cited 1× · see citing works
“procedure described in [15] and the pulse2percept software [16] to perform image registration and extract the location of the implant and each electrode with respect to the fovea (Fig.”
Self-citationThe Sensorium competition on predicting large-scale mouse primary visual cortex activityPreprint
Willeke, K. F., Fahey, P. G., Bashiri, M., Hansel, L., Burg, M. F., Blessing, C., Cadena, S. A., Ding, Z., Lurz, K.-K., Ponder, K., et al. (2022) · arXiv (Cornell University) · doi:10.48550/arxiv.2206.08666 · Cited by 22 · cited 10× · see citing works
“Note that FEVE0 ≡ FEVE from Eq.”
InfluentialA model of ganglion axon pathways accounts for percepts elicited by retinal implants
Beyeler, M., Nanduri, D., Weiland, J. D., Rokem, A., Boynton, G. M., & Fine, I. (2018) · doi:10.1101/453035 · Cited by 13 · see citing works
Self-citationA Patient-Specific Computational Framework for the Argus II Implant
Finn, K. E., Zander, H. J., Graham, R., Lempka, S. F., & Weiland, J. D. (2020) · IEEE Open Journal of Engineering in Medicine and Biology · doi:10.1109/ojemb.2020.3001563 · Cited by 13 · see citing works
Long term positional stability of the Argus II retinal prosthesis epiretinal implant
Ghani, N., Bansal, J., Naidu, A., & Chaudhary, K. M. (2023) · BMC Ophthalmology · doi:10.1186/s12886-022-02736-w · Cited by 9 · cited 2× · see citing works
“We therefore had to assume that the location of the array stays stable over time, which is supported by a recent study highlighting the long-term stability of the Argus II implant [14].”
Hu, Z.-Y., & Beyeler, M. (2021) · International IEEE/EMBS Conference on Neural Engineering · doi:10.1101/2021.03.07.21253092 · Cited by 9 · cited 2×
“Part of this work was previously presented in [8].”
Self-citationHu, Z., & Beyeler, M. (2021) · International IEEE/EMBS Conference on Neural Engineering · doi:10.1109/ner49283.2021.9441194 · Cited by 7 · see citing works
Self-citationLong term positional stability of the Argus II Retinal Prosthesis epiretinal implantPreprint
Ghani, N., Bansal, J., Naidu, A., et al. (2022) · Research Square · doi:10.21203/rs.3.rs-1650046/v1 · Cited by 2 · see citing works
(2017) · doi:10.9728/dcs.2024.25.7.1847 · Cited by 2 · see citing works
Flagged first, then by citations
Cited by
Sorted by citations
Human-in-the-Loop Optimization for Deep Stimulus Encoding in Visual Prostheses
Granley, J., Fauvel, T., Chalk, M., & Beyeler, M. (2023) · PubMed · doi:10.48550/arxiv.2306.13104 · Cited by 8 · see citing works
Self-citationAxonal stimulation affects the linear summation of single-point perception in three Argus II users
Hou, Y., Nanduri, D., Granley, J., Weiland, J. D., & Beyeler, M. (2024) · Journal of Neural Engineering · doi:10.1088/1741-2552/ad31c4 · Cited by 14 · see citing works
Self-citationFerhi, L. A., Amar, M. B., Choubani, F., & Bouallègue, R. (2024) · Frontiers in Artificial Intelligence · doi:10.3389/frai.2024.1397388 · Cited by 14 · see citing works
Patient-specific computational models of retinal prostheses
Kish, K. E., Yuan, A., & Weiland, J. D. (2023) · Scientific Reports · doi:10.1038/s41598-023-49580-6 · Cited by 7 · see citing works
Visual Prostheses in the Era of Artificial Intelligence Technology
Sarbout, I., Gungor, A., Ounissi, M., Zaher, S., Ptito, M., Kupers, R., Racoceanu, D., & Milea, D. (2025) · Eye and Brain · doi:10.2147/eb.s524322 · Cited by 3 · see citing works
“95 Another approach consisted in using human-in-the-loop optimization to improve the perceived image in a simulated prosthetic vision setting.”
Skaza, J., Murlidaran, S., Varshney, A., Wen, Z., Eckstein, M. P., & Beyeler, M. (2025) · bioRxiv (Cold Spring Harbor Laboratory) · doi:10.1101/2025.06.23.660990
“Simulations also assume stable electrode performance and consistent percepts, whereas clinical data reveal substantial variability and adaptation [29, 26].”
Self-citationSadeghi, R., & Beyeler, M. (2025) · arXiv.org · doi:10.48550/arxiv.2502.06672
“1), which can be attributed to factors such as electrode-retina distance, impedance variations, and localized retinal damage [3], [7].”
Self-citationAxonal stimulation affects the linear summation of single-point perception in three Argus II users
Hou, Y.-C., Nanduri, D., Granley, J., et al. (2023) · medRxiv · doi:10.1101/2023.07.21.23292908
“…electrode-retina distance and perceptual thresholds has been well established in the literature (Mahadevappa et al., 2005; de Balthasar et al., 2008; Ahuja et al., 2013; Pogoncheff et al., 2023), research examining the effect of these parameters on the shape of elicited phosphenes has been limited.”
Self-citation
- Authors
- Pogoncheff, G., Hu, Z., Rokem, A., & Beyeler, M.
- Affiliations
- University of California, Santa Barbara (US)
- University of Washington (US)
- Corresponding author
- Pogoncheff, Galen
- Venue
- Journal of Neural Engineering
- Publisher
- openRxiv
- Published
- 2023-02-10
- Volume
- 21
- Issue
- 2
- Pages
- 026009-026009
- ISSN
- 1741-2552 (ISSN-L)
- PMC
- PMC9934792
- OpenAlex
- W4392564368
- Cited by
- 9
- References
- 24
- Citations by year
- 2026: 2 · 2025: 2 · 2024: 3 · 2023: 2
- FWCI
- 1.22
- Type
- journal article
- Language
- en
- Open access
- green
- doi.org
- Licence
- cc-by
- Locations
- Journal of Neural Engineering (publishedVersion)
- medRxiv (acceptedVersion, cc-by)
- PubMed (publishedVersion)
- PubMed Central (acceptedVersion, cc-by)
- PubMed Central (submittedVersion, cc-by)
- eScholarship (California Digital Library) (submittedVersion, cc-by)
- Funders
- University of Washington
- National Institutes of Health
- Institute on the Environment, University of Minnesota
- National Eye Institute (R00EY029329)
- Topics
- Life Sciences › Neuroscience › Cellular and Molecular Neuroscience › Neuroscience and Neural Engineering
- Life Sciences › Neuroscience › Cognitive Neuroscience › Tactile and Sensory Interactions
- Health Sciences › Medicine › Anesthesiology and Pain Medicine › Pain Management and Treatment
- Keywords
- retinal prosthesis, perceptual sensitivity, perceptual thresholds, electrode deactivation, explainable AI, stimulation parameters, age effects, visual outcome prediction
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.
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
