Visual cortical thickness increases with prolonged artificial vision restoration

Abstract

The Argus II retinal prosthesis restores visual perception to late blind individuals. It has been shown that structural changes occur in the brain due to late-onset blindness, including cortical thinning in visual cortical regions. Following vision restoration, it is not yet known whether visual regions are reinvigorated and regain a normal cortical thickness or retain the diminished thickness from blindness.

We evaluated the cortical thicknesses of 10 Argus II participants, 10 blind participants, and 13 sighted participants. The Argus II patients on average had a thicker left cuneus cortex and lateral occipital cortex relative to the blind patients. The duration of the Argus II use significantly partially correlated with thicker visual cortical regions in the left hemisphere. Furthermore, in the 2 case studies (scanned before and after implantation), the patient with longer device use (44.5 months) had an increase in the cortical thickness of visual regions, whereas the shorter-using patient did not (6.5 months).

Overall, the Argus II participants’ cortical thickness was on average significantly rejuvenated in 2 higher visual regions, and participants using the implant for a longer duration had thicker visual regions. This research raises the possibility of structural plasticity reversing visual cortical atrophy in vision restoration participants.

Introduction

Extensive and complex damage to retinal cells, vasculature, and neural networks has previously prevented the restoration of visual perception to those blinded by degenerative retinal disease. However, modern advances in biomedical engineering, as well as molecular and cell biology, over the past few decades have facilitated the innovation of emerging therapeutics and implants to restore functional vision to the visually impaired. These therapeutics and implants include biomedical devices such as retinal and cortical prostheses (Weiland et al. 2011; Zrenner et al. 2011; Humayun et al. 2012; Zhou et al. 2013; Zrenner 2013; Luo and Da Cruz 2014; Luo et al. 2016; Luo and da Cruz 2016; Niketeghad and Pouratian 2019; Beauchamp et al. 2020), as well as biomolecular therapies, such as gene therapies (Apte 2018; Lam et al. 2019; Mowad et al. 2020), optogenetic therapies (Baker and Flannery 2018; Simunovic et al. 2019; Ferrari et al. 2020), and stem cell therapies (Kashani et al. 2018; Roska and Sahel 2018). Retinal prostheses have been one of the approaches most studied and realized, with hundreds of people already implanted with prostheses worldwide.

Retinal prostheses restore low-resolution visual perception to the blind by stimulating viable retinal cells according to a video stream from a head-mounted camera (Fig. 1) (Zhou et al. 2013; Luo and da Cruz 2016). In this paper we study the Argus II retinal prosthesis, which was manufactured by Second Sight Medical Products and has been approved by the FDA for implantation in one eye of individuals with light perception or less due to Retinitis Pigmentosa (RP). The Argus II device has been implanted in approximately 350 visually impaired individuals worldwide (2019).

Fig. 1.

Argus II device external components. An image of the Argus II retinal prosthesis external components including the pair of glasses with a head mounted camera and the visual processing unit. The implanted internal components are not included in this image.

The Argus II device consists of 60 electrodes (60 pixels), which generate ultra-low resolution vision with the ability to determine the direction of motion, match auditory and visual stimuli, identify basic shapes, and perform object localization (Humayun et al. 2012; Kotecha et al. 2014; Luo et al. 2014; Stronks and Dagnelie 2014; Stiles et al. 2021). However, Argus II artificial visual perception is quite different than natural vison. In particular, prosthesis-driven perception is highly variable across the patient population and can require substantial effort and attention to perform basic tasks at a slow rate (compared to low vision and sighted individuals) (Zhou et al. 2013; Luo et al. 2016; Luo and da Cruz 2016). These challenges could be due to the elongated and curved shape of the phosphenes (pixels) generated by electrical stimulation (Beyeler et al. 2019), perceptual fading (Fornos et al. 2012; Avraham et al. 2021), deterioration of retinal circuitry during blindness (Marc and Jones 2003), the abnormal stimulation of retinal ganglion cells by electrodes (Beyeler et al. 2019), and/or visual cortical changes during blindness (Pascual-Leone and Hamilton 2001; Sadato et al. 2004). The cortical changes during blindness can include both functional changes, such as the crossmodal repurposing of dormant visual regions with auditory or tactile processing (Sadato et al. 2004; Amedi et al. 2005; Amedi et al. 2007), and structural changes, such as the atrophy of visual white matter pathways and gray matter cortical thicknesses (Leporé et al. 2010; Burge et al. 2016; Machado et al. 2017). Both the functional and structural changes of the visual cortical neural network during decades of late blindness may limit the behavioral outcomes of Argus II participants. In addition, the ability for the visual cortex to reorganize and adapt to restored visual input may have implications for Argus II outcomes.

In this paper, we study the structural changes of the gray matter cortical thicknesses in Argus II retinal prosthesis participants with late-stage RP. Late-stage RP causes at a minimum severe constriction of peripheral vision; however, it can also progress to complete loss of visual perception with only light perception remaining (as with the Argus II device candidates). Individuals with late-stage RP, similar to most late blind, have been shown to have significant changes to the structure of the visual cortex due to vision loss (Cunningham et al. 2015a; Ferreira et al. 2017; Machado et al. 2017; Sanda et al. 2018; Castaldi et al. 2019). In particular, late-stage RP participants have been found to have significant thinning of primary visual cortex when compared to sighted controls (Sanda et al. 2018), which is hypothesized to be due to atrophy that progresses from the damaged retina, extending through the visual pathway to visual cortical regions (ie trans-synaptic neurodegeneration or Wallerian degeneration) (Machado et al. 2017). RP participants have also been shown to have a reduction in visual cortex total gray matter that negatively correlates with visual function. In other words, participants with more severe vision loss were found to have more gray matter loss in visual cortical regions (Machado et al. 2017).

In this study, we investigate whether vision restoration by prostheses such as the Argus II device can reverse the cortical structural atrophy generated by an extended period of blindness due to late-stage RP. We evaluate this question by comparing the visual cortical thickness of 10 Argus II retinal prosthesis participants to 10 late-stage RP participants (blind controls) and 13 sighted controls (both age-matched). We will also evaluate the visual cortical thicknesses before and after implantation in 2 Argus II case studies and at multiple timepoints following implantation in a third Argus II case study.

Materials and methods

Participants

Ten Argus II participants (7 male, 3 female) with an average age of 61 years old participated in this research study at the University of Southern California (USC) (Table 1). The Argus II participants had the device implanted for an average of 51.60 months (4.30 years) and experienced substantial vision loss for an average of 32.22 years (note: 1 participant did not report their duration of blindness). All of the Argus II participants had light perception or less in both eyes when the Argus II device was turned off, and the Argus II device was implanted in the left eye in 3 participants, and the right eye in 7 participants.

Table 1.

Participant demographic information.

Group	Subject ID	Age (years)§	Gender	Duration blind (years)	Duration with Argus II (months)	Vision left (OS)	Vision right (OD)
Argus II cohort	A1	53	F	26	27.5	LP	NLP†
	A2	61	F	16	43.5	LP	NLP†
	A3	46	M	20	19.5	LP	LP†
	A4	61	F	‐	6.5	LP or less†	LP or less
	A5	65	M	26	23	LP	NLP†
	A6	76	M	38	19	NLP†	NLP
	A7	44	M	23	121	LP	LP†
	A8	62	M	39	99.5	LP	LP†
	A9	76	M	56	112	LP	NLP†
	A10	66	M	46‡	44.5	NLP†	NLP
Blind controls	B1	70	F	52	‐	LP or less	LP or less
	B2	51	F	33	‐	LP or less	LP or less
	B3	60	F	‐	‐	LP or less	LP or less
	B4	61	M	30‡	‐	NLP	NLP
	B5	57	M	24	‐	HM	HM
	B6	55	M	51	‐	20/80−2*	20/50+2*
	B7	44	F	37	‐	LP	LP
	B8	24	M	17	‐	HM	20/1600
	B9	48	M	‐	‐	20/800*	20/300+1*
	B10	40	M	35	‐	HM	HM
Sighted controls	S1	44	M	‐	‐	20/20	20/20+1
	S2	64	F	‐	‐	20/50−3	20/40
	S3	66	F	‐	‐	20/20−3	20/20−3
	S4	63	M	‐	‐	20/60+1	20/50+2
	S5	44	F	‐	‐	S	S
	S6	41	M	‐	‐	S	S
	S7	34	M	‐	‐	S	S
	S8	61	F	‐	‐	S	S
	S9	61	F	‐	‐	S	S
	S10	65	M	‐	‐	S	S
	S11	46	M	‐	‐	S	S
	S12	53	F	‐	‐	S	S
	S13	76	M	‐	‐	S	S

M, male; F , female; LP, light perception; NLP, no light perception; HM, hand motion; LP or less, vision is not tested by research team but the patient qualified for the Argus II device, which requires light perception or less; S, normally sighted individual from the Lifespan HCP-A study or the USC study without acuity testing provided; ‐ = Information is either not available or not applicable.

^*Visual field diameter measured by MAIA microperimetry for patient B6 and B9 was 2° or less in both eyes (Supplementary Figs S1 and S2).

^†The eye of the Argus II implantation.

^‡B4 and A10 are the same individual 5 years apart. The duration blind is different between these 2 timepoints by more than 5 years because the Argus II patient duration blind was based on a questionnaire and the blind patient duration blind was based on a clinical report; these 2 measures inquired about blindness onset in different but consistent ways within each cohort.

^§The age of participants was not significantly different between the Argus II patients and 2 control groups (Argus II vs. blind controls: P = 0.08; Argus II vs. sighted controls: P = 0.26).

Argus II participants A4 and A10 (included in the Argus II participant group) were also scanned before device implantation. This pre-implantation data from patient A4 and A10 are included in the blind cohort (detailed below), as patient B3 and B4, respectively. In addition, we also compared the pre-implantation and post-implantation cortical thicknesses for these 2 participants as 2 separate case studies in the results section without statistical analyses.

One other Argus II participant also participated in a longitudinal case study post-implantation. Participant A2 was scanned at 43.5, 57, and 106 months after Argus II implantation. At the first 2 scans, she was actively using the Argus II device; however, at the last scan at 106 months, she had not been using the Argus II device for 36 months. The first scan of participant A2 (at 43.5 months) is included in the group level analysis, and a case study comparison between the 3 scans is also included in the results section without statistical analyses.

Ten advanced RP control participants participated in this research study at USC (Table 1). The advanced RP participants are called the “blind control group” throughout the paper. The blind control group had an average age of 51 years at the time of the experiment and consisted of 6 male and 4 female individuals. The blind control group all had severe vision loss with a range of acuity between no light perception and 20/50+2 (Table 1). The only participant with acuity above 20/300+1 was participant B6, who had a visual acuity of 20/80−2 and 20/50+2 in left and right eyes, respectively. Participant B6 was classified as severe RP and included in the blind group because he had a visual field of 2° and a low level of light detection in both eyes as measured with MAIA microperimetry (Supplementary Fig. S1). Participant B9 (visual acuity: 20/800 in the left eye and 20/300 + 1 in the right eye; Table 1), also had a visual field of 2° as measured with MAIA microperimetry (Supplementary Fig. S2).

Note: 4 of the blind control participants (B1, B2, B3, and B4) qualified for the Argus II implant. Two of these participants received the implant and were scanned after implantation as participant A4 and A10 (detailed above). The other 2 participants either did not receive the implant or did not return for a second scan following implantation.

Thirteen sighted control individuals participated in this research study. Seven of the sighted controls (Participants S1 through S7) had data collected at USC with the same protocol as the Argus II participants and blind control participants detailed above. Due to the small size of the USC sighted cohort, 6 of the sighted controls (participants S8 through S13) were included from the Lifespan Human Connectome Project Aging (HCP-A) study (Bookheimer et al. 2019). The HCP-A participants were chosen from that database to match the age and gender of the first 6 Argus II participants (in order from S8 through S13 the HCP-A participants included are HCA61, HCA71, HCA73, HCA76, HCA79, and HCA88). A comparison between the 2 groups (USC sighted controls and HCP-A sighted controls) did not show a significant difference (Supplementary Fig. S3) except for left pericalcarine cortex (P = 0.04) and left lingual gyrus (P = 0.04). The sighted control group (including participants from USC and HCP-A) had an average age of 55.23 years old and consisted of 7 males and 6 females (Table 1).

The neuroimaging data for the Argus II participants (Participants A1 through A10), the blind participants (Participants B1 through B10), and 7 of the sighted participants (Participants S1 through S7) were collected at USC and approved by the USC Institutional Review Board (Protocols HS-16-00090 and HS-16-00441). The participants all gave written informed consent before performing the experiments.

Argus II retinal prosthesis system

The Argus II retinal prosthesis system was manufactured by Second Sight Medical Products (discontinued). The device consists of an implanted electrode array and coil, and the external glasses and visual processing unit (Fig. 1). The array has 60 electrodes, which electrically stimulate the retinal surface. Visual information is initially captured by a camera on the pair of glasses that sends this information to the video processing unit to convert the images into stimulation parameters. The signal is then sent back to the glasses via a cable and communicated into the eye with a glasses-mounted coil, which transmits it to an internal coil and then the electrode array. The Argus II was approved for individuals with RP and a visual acuity of light perception or less. Outside this indication, it has been implanted in individuals with dry age-related macular degeneration (AMD) in a clinical trial.

Following device implantation and activation, Argus II implantees are offered rehabilitation training (often in their home environment) to facilitate the learning of basic visual activities and functions. Seven of the Argus II participants in this study participated in the Second Sight-provided training (1 participant did not provide this information). Additional information about the training of the Argus II participants is detailed in Supplementary Table S1. Following directed rehabilitation training, implantees can use the device in the activities of daily living. The Argus II participants scanned were active users of the device, 9 participants self-reported their frequency of use, which ranged from every day to once per month (frequency averaged over total time since implant) (Supplementary Table S1).

USC and HCP-A MRI scanning protocols

The USC MRI scans were performed on a 3-Tesla Siemens Prisma MRI scanner with a 32-channel Siemens head coil. For each participant a T1-weighted anatomical image with magnetization-prepared rapid gradient-echo (MPRAGE) was acquired in axial orientation (TR = 2400 ms, TE = 2.22 ms, flip angle = 8, FOV = 256 × 256 mm², voxel size 0.8-mm isotropic voxels, number of slices = 208, acceleration factor (GRAPPA) = 2). The data collection followed the Human Connectome Project LifeSpan protocol (VD13D). The sighted control data for participants S8 through S13 was derived from the HCP-A database (Bookheimer et al. 2019). The HCP-A data was acquired on a 3-Tesla Siemens Prisma MRI scanner with a 32-channel head coil. The T1-weighted anatomical image was collected with multiecho MPRAGE (TR/TI = 2500/1000, TE = 1.8/3.6/5.4/7.2 ms, flip angle = 8, sagittal FOV = 256 × 240 × 166 mm, voxel size 0.8-mm isotropic voxels, 4 echoes per line of k-space, matrix = 320 × 300 × 208 slices, 7.7% slice oversampling, pixel bandwidth = 744 Hz/pixel, acceleration factor (GRAPPA) = 2). Motion-induced re-acquisition were allowed for up to 30 TRs (Harms et al. 2018).

MRI data processing details

The T1-weighted structural images were processed in Freesurfer 7.1.1 software using the recon-all function, with a fully automated directive (−all). (Desikan et al. 2006; Klein and Tourville 2012). The longitudinal case studies were also evaluated using a longitudinal analysis with the recon function, which allows 2 scans to share a structural template and then be directly compared. The analysis first performed a fully automated directive (recon-all) at each timepoint, then evaluated the 2 timepoints with a -base directive (recon-all -base) to create a within subject template, and lastly each timepoint was processed with the longitudinal analysis directive (recon-all -long) (https://surfer/nmr.mgh.harvard.edu/fswiki/FsTutorial/LongitudinalTutorial). The longitudinal analysis makes a common template for 2 scans; therefore, the 2 case studies scanned before and after implantation (A4 and A10) compared these 2 scans, whereas the case study with 3 scan timepoints (A2) compared each scan to the scan at the previous timepoint (Supplementary Figs. S9 and S10).

Note: cortical thickness measured through T1 images has limitations and is different than the anatomical cortical thickness measured postmortem. It is likely influenced by myelination and blood flow and therefore is closer to an “apparent” cortical thickness (Olivo et al. 2022).

Four visual regions of interest were compared between participant groups (Argus II participants, blind controls, and sighted controls): pericalcarine cortex, cuneus cortex, lingual gyrus, and lateral occipital cortex. The regions of interest were selected to include all occipital lobe regions as delineated by the Desikan–Killiany atlas. The left and right cortical regions were evaluated separately.

The segmentation of the thalamic nuclei was performed with the segmentThalamicNuclei.sh function in Freesurfer 7.1.1. The left and right lateral geniculate nucleus (LGN) volumes were retrieved from the thalamic-nuclei.lh.v12.T1 and the thalamic-nuclei.rh.v12.T1 files, respectively. The left and right LGN volumes were normalized by the brain segmentation volume without ventricles (aseg file).

Normalization was performed by dividing each cortical thickness or region volume by a participant’s average cortical thickness or hemispheric brain volume. Normalization was used in order to remove the cross-subject variation due to differences in total brain volume. As hypothesized, the non-normalized data showed a similar pattern of cortical thickness but with additional interparticipant noise (Supplementary Fig. S4). This normalization follows the standard practice in the cortical thickness literature (Cunningham et al. 2015a; Jiang et al. 2015; Aguirre et al. 2016; Ferreira et al. 2017; Sanda et al. 2018).

Statistical analyses and data visualization

The cortical thickness data was evaluated for normality using the Shapiro–Wilk and Shapiro–Francia tests (BenSaïda 2014) and failed to be classified as normal. Therefore, the Kruskal–Wallis nonparametric one-way ANOVA analysis was performed for each region of interest. If the one-way ANOVA showed a significant variation between participant groups (P < 0.05), group comparisons were made. The individual group pairs were compared using the multcompare function (Fisher’s least significant difference procedure). The multcompare function based its analysis on a one-way ANOVA for the cortical region of interest comparing the 3 participant groups (this prevented comparisons between the cortical regions, which was not of interest). The visualization of the Desikan–Killiany atlas was performed in FreeView.

Partial correlations were performed in MATLAB using the partialcorr function. A partial correlation is used when a third variable could confound or alter the relationship between the primary 2 variables. In our case, as participants have a longer period of blindness or device use, they also age, and age can also impact brain structure. Therefore, age was identified as a confounding factor for the duration of blindness and device use and statistically accounted for by using a partial correlation. Overall, correlation analyses were performed between cortical thickness and the participants’ duration of prosthesis use (calculated as the period between participants’ self-reported date of implantation and the date of testing), the participants’ duration of blindness (self-reported), and the Argus II participant functionality [a shape matching task, methods for this task are detailed in Stiles et al. (2022)] while controlling for participant age.

Results

Group level analyses

Left hemisphere

The left pericalcarine cortex did not have significant differences among the participant groups [H (2, n = 33) = 2.13, P = 0.34] (Fig. 2; Supplementary Fig. S5).

Fig. 2.

Visual region cortical thicknesses. Panel (A) shows the parcellation of the occipital regions with the Desikan–Killiany atlas in an example Argus II participant, A1. Note: the pericalcarine cortex (light blue) can be found between the cuneus cortex and lingual gyrus as a sulcus (inward cortical fold). Panel (B) shows the cortical thicknesses for the 4 visual areas highlighted in Panel (A) in the 3 participant groups: the blind participants, the Argus II participants, and the sighted controls (detailed in Table 1). The cortical thickness measures are normalized by the average cortical thickness in the same hemisphere for each participant. Purple stars indicate significant differences between participant groups (P < 0.05). The full length of each error bar represents 1 standard deviation.

The left cuneus cortex had significant variation across the participant groups [H (2, n = 33) = 7.41, P = 0.02]. The Argus II participants and the sighted participants had on average a significantly thicker left cuneus cortex than the blind participants (Argus II participants vs. blind controls: P = 0.03; sighted controls vs. blind controls: P = 0.01) (Fig. 2; Supplementary Fig. S5). The Argus II participants did not have a significantly different cortical thickness than the sighted controls (P = 0.74) (Fig. 2; Supplementary Fig. S5).

The left lingual gyrus did not have significant differences among the participant groups [H (2, n = 33) = 5.62, P = 0.06] (Fig. 2; Supplementary Fig. S5).

The left lateral occipital cortex had significant variation across the participant groups [H (2, n = 33) = 7.16, P = 0.03]. The Argus II participants and the sighted participants were found to have a significantly thicker left lateral occipital cortex than the blind participants (Argus II participants vs. blind controls: P = 0.04; sighted controls vs. blind controls: P = 0.01) (Fig. 2; Supplementary Fig. S5). The Argus II participants’ cortical thickness of left lateral occipital cortex was not significantly different than the sighted controls (Argus II participants vs. sighted controls: P = 0.68) (Fig. 2; Supplementary Fig. S5).

The left LGN volume results are reported in Supplementary Fig. S6.

Right hemisphere

The right hemisphere did not have any significant differences between participant groups in the pericalcarine cortex [H (2, n = 33) = 2.07, P = 0.36], cuneus cortex [H (2, n = 33) = 5.22, P = 0.07] or lateral occipital cortex [H (2, n = 33) = 5.09, P = 0.08] (Fig. 2; Supplementary Fig. S5).

However, the right lingual gyrus had significant variation across the participant groups [H (2, n = 33) = 7.34, P = 0.03]. The sighted participants had on average a significantly thicker right lingual gyrus than the Argus II participant group and the blind control group (sighted controls vs. Argus II participants: P = 0.04; sighted controls vs. blind controls: P = 0.01) (Fig. 2; Supplementary Fig. S5). The Argus II participants’ cortical thickness of right lingual gyrus was not significantly different from the blind controls (P = 0.75) (Fig. 2; Supplementary Fig. S5).

The right LGN volume results are reported in Supplementary Fig. S6.

Correlations between cortical thickness and demographics

The duration of prosthesis use (n = 10) was significantly positively partially correlated with the average cortical thickness of the 4 visual regions of interest of the left hemisphere (rho = 0.83, P = 6.2 × 10⁻³) but not the right hemisphere (rho = 0.29, P = 0.45) (controlling for age) (Fig. 3). To more precisely evaluate this correlation, partial correlation analyses for each visual region in the left hemisphere were also performed. The cortical thickness of the left pericalcarine cortex (rho = 0.68, P = 0.04) and cuneus cortex (rho = 0.82, P = 6.8 × 10⁻³) significantly partially correlated with the duration of prosthesis use (controlling for age).

Fig. 3.

Correlation between visual cortical thickness and the duration of Argus II device use. Figure 3 shows a correlation plot for the Argus II participants (n = 10) between the visual cortical thickness and the duration of Argus II device use. Each color represents a different visual region of interest including the left and right hemispheres of the pericalcarine cortex, cuneus cortex, lingual gyrus, and lateral occipital cortex. Each square or dot is a data point for 1 subject in that visual region of interest.

However, an Argus II functionality measure (shape matching task derived from Stiles et al. (2022) detailed in the Materials and methods section) and the participants’ duration of blindness did not partially correlate (controlling for age) with the average cortical thickness of the 4 visual regions of interest (Shape Task, n = 8: left hemisphere: rho = −0.13, P = 0.77; right hemisphere: rho = 0.18, P = 0.69) (duration of blindness, n = 9: left hemisphere: rho = 0.62, P = 0.18; right hemisphere: rho = −0.19, P = 0.65).

Individual longitudinal analyses

Pre- and Post-implantation scans (2 case studies)

Participants A4 and A10 were scanned before and after implantation with the Argus II device, permitting a direct within-participant comparison of visual cortical thickness (the left and right LGN volumes are reported in Supplementary Fig. S7). Participant A4 was scanned 6.5 months (0.54 years) following Argus II implantation and had an average reduction in visual cortical thickness of 1.9% relative to the pre-implantation scan (Fig. 4a). Participant A10 was scanned 44.5 months (3.71 years) following the Argus II implantation and had an average increase in visual cortical thickness of 2.2% relative to the pre-implantation scan (Fig. 4b). Participant A10 showed the largest increases in left lingual gyrus (5.0%), the left cuneus cortex (3.5%), and the right lingual gyrus (4.3%) thickness. Longitudinal analyses were also performed (details in Materials and methods), which uses a common structural template between the 2 scans. While these results were consistent with differences reported above in general, they did have smaller effect sizes (Supplementary Fig. S9).

Fig. 4.

Individual pre and post implantation visual cortical thicknesses. Figure 4 shows the visual cortical thicknesses for 2 case studies, A4 and A10, which have scans before and after the implantation of the Argus II device (details in Table 1). Participant A4 was scanned 6.5 (0.54 years) after Argus II implantation and Participant A10 was scanned 44.5 months (3.71 years) after Argus II implantation.

Multiple post-implantation scans (1 case study)

Participant A2 was scanned 2 additional times following the first scan at 43.5 months (3.63 years) post-implantation. The second scan was at 57 months (4.75 years) post-implantation with the device still in active use. The final scan was at 106 months (8.83 years) post-implantation but 36 months (3 years) after ceasing to use the device. The cortical thicknesses in the left and right pericalcarine cortex, cuneus cortex, and lingual gyrus decreased over 1 year of device use (between 3.63 years and 4.75 years). In all these regions except the left pericalcarine cortex, the cortical thickness rebounded following the years of disuse (Supplementary Fig. S8A). The cortical thickness in the lateral occipital cortex (left and right hemispheres) increased over 1 year of Argus II use (similar to the increase in thickness in the group level analysis). The increased cortical thickness in the left and right lateral occipital cortices was maintained after disuse of the Argus II device. The left and right LGN volumes are reported in Supplementary Fig. S8B. Longitudinal analyses were also performed (details in Materials and methods), which uses a common structural template between pairs of scans. While these results were largely consistent with differences reported above, they did have smaller effect sizes (Supplementary Fig. S10).

Discussion

Overview

Multiple early visual regions had significantly reduced cortical thickness or volume in the blind control group relative to the sighted control group, including the left cuneus cortex, the left lateral occipital cortex, the right lingual gyrus, and the right LGN. Similar to the blind control group, the Argus II participant group also had a significantly thinner cortical thickness in the right lingual gyrus and a significantly reduced volume in the left and right LGN relative to the sighted control group. However, the Argus II participant group had a significantly thicker left cuneus cortex and left lateral occipital cortex relative to the blind cohort, and these thicknesses were similar in magnitude to the sighted control group. The Argus II participants’ cortical thickness across all left visual regions significantly partially correlated with the duration since implantation. Furthermore, the pre- and post-implantation case studies show no cortical thickening with a half year of use, but thickening with nearly 4 years of use. In another case study, the participant had thicker left and right lateral occipital cortices between 3.63 and 4.75 years of device use, and a maintenance of that increased thickness following 3 years of disuse.

These results show that the Argus II participants can have a significant restoration of visual gray matter with artificial vision, especially following prolonged use and in visual cortical regions. As would be expected with structural changes in the brain, this rejuvenation of thickness increased with longer device usage and appeared to require years of rehabilitated vision to be fully instantiated.

Vision loss and cortical thinning

Visual impairment due to retinal disease in the late blind has been shown to reduce the cortical thickness of visual regions (Cunningham et al. 2015a; Ferreira et al. 2017; Machado et al. 2017; Sanda et al. 2018; Castaldi et al. 2019). Cortical gray matter loss in participants with retinal disease is hypothesized to originate in the retina and gradually progress up the visual pathway via trans-synaptic neurodegeneration called Wallerian degeneration (Machado et al. 2017). At the cellular level, gray matter loss is hypothesized to be due to “neuronal apoptosis, variations in cortical myelination, alterations of the synaptic complexity, or a summation of these events” (Sanda et al. 2018, p. 3474) (Wagstyl et al. 2015; Zilles and Amunts 2015; Burge et al. 2016; Sanda et al. 2018).

In this paper, we investigated late-blind RP participants, which have been shown to have thinner cortical thickness in visual regions. In particular, Castaldi et al. (2019) found significant thinning of V1 in RP participants compared to sighted controls; however, this difference may be due to the younger age of the sighted controls. In Sanda et al. (2018) RP participants with tunnel vision (10° to 20°) were found to have thinner cortical thickness in early visual cortex, V1 and V2, as well as V3d and V4 (Sanda et al. 2018). Machado et al. (2017) studied the total gray matter volume in visual cortical regions and found that the left and right calcarine sulci, left and right lingual gyri, left and right cuneus cortices, and right superior gyrus of the occipital lobe were significantly reduced in volume in RP participants relative to sighted controls. In contrast to these results, Ferreira et al. (2017) found no significant differences between the cortical thickness of the RP participants they evaluated and their sighted controls. Interestingly, when they only evaluated the region of primary visual cortex with spared vision, they found that RP participants with less than 15° visual field had a thicker visual cortex than the RP participants with more than a 15° visual field. This thickening of the more impaired participants may be due to compensation in regions of spared vision. This contrast of thinning in regions of impairment and thickening in regions of spared vision may contribute to the intersubject variability in RP participants.

The late-stage RP participants studied in this paper (ie the blind control group) had a severe vision loss, ranging from complete blindness (no light perception) to 2° of tunnel vision remaining. Similar to results in other studies, we found significant degradation of gray matter in these RP participants from visual subcortical regions (such as LGN) to visual cortical regions (such as the cuneus cortex). An interesting difference between this study and Sanda et al. (2018) is that our RP cohort did not have a significantly thinner pericalcarine cortex (or early visual cortex) than the sighted controls. While the RP participants in our study had on average a substantially thinner pericalcarine cortex relative to the sighted control participants, this difference was not significant due to the large variability within the RP group. This high variability in the cortical structure and function of low vision and blind individuals is common in the neuroscience literature (Cunningham et al. 2015b; Ferreira et al. 2017) and could in this case be caused in part by the conflicting processes of thinning and thickening of V1 during the progression of RP (as highlighted above). Overall, the RP cohort in this study had significant gray matter loss (cortical thickness and subcortical volume) in multiple visual regions, which is expected due to the severity and long duration of vision loss in the participants tested.

Cortical thickness changes with vision restoration

Cortical thickness has been previously evaluated in 2 vision restoration populations: Argus II, and cataract participants. The 2 previous Argus II studies, which evaluated cortical thickness, were case studies (Cunningham et al. 2015a; Stanga et al. 2021). Cunningham et al. (2015a) evaluated the V1 cortical thickness of 2 Argus II participants with RP (duration of prosthesis use were 1.5 and 3.75 months) relative to late-stage RP controls (n = 9) and sighted controls (n = 9). They found that the 2 Argus II participants’ V1 cortical thicknesses were within the ranges of the blind and sighted control groups and argued that MRI imaging was feasible with the Argus II retinal prosthesis. The Cunningham participants were different than our participant group in 2 key aspects, Cunningham et al. (2015a) did not have pre-implantation and post-implantation within-subject comparisons, and their participants used the Argus II device for less than 4 months. Our analyses included 2 pre- and post-implantation case studies, 1 longitudinal post-implantation case study, and 7 additional Argus II participants with a range of use from 19 months to 121 months.

Stanga et al. (2021) measured the cortical thickness of V1 and V2 in 1 AMD participant before and 13 months after implantation with the Argus II device. The cortical thickness of the Argus II participant’s V1 and V2 increased after implantation but was still below the age-matched sighted controls (n = 8) average cortical thickness. Stanga et al. (2021) studied an AMD participant with potentially substantial remaining visual perception; therefore, their result is not highly comparable to our study in RP participants with light perception or less. However, they do show visual region rejuvenation with Argus II device use, which is generally consistent with our results.

Guerreiro et al. (2015) studied cortical thickness changes in participants with sight recovery due to congenital cataract surgery. Cataract surgery was performed in these participants between 5 and 24 months from birth and therefore within the visual critical period. Early-blind participants with vision loss during the critical period have been previously found to have thicker early visual cortices, which is hypothesized to be due to an absent or reduced cortical pruning phase of visual development (Guerreiro et al. 2015). Guerreiro found that early-blind participants with vision restoration following cataract extraction maintained this thickened visual cortex relative to sighted controls. Therefore, the decades of restored visual perception did not reverse the changes in cortical architecture due to blindness during the critical period. The critical period is a developmental phase with strong structural plasticity, and it is followed by adulthood with more limited plasticity. Our participant group were late blind and therefore did not have vision loss during the critical period but rather during adulthood. It is therefore possible that while the structural changes due to impairment during the critical period cannot be reversed in individuals with congenital cataracts, our results show a potential to partially reverse changes that can occur in adulthood in Argus II participants.

Our group results show that 2 later visual regions (the left cuneus cortex and lateral occipital cortex) had significant rejuvenation in Argus II participants that was equivalent to the sighted controls. Furthermore, correlation analyses between visual cortical region thickness and Argus II participants’ duration of use, as well as case studies fortify our group result by showing more rejuvenation with longer device use. If it is the case that visual cortical thickness is increased with vision restoration, we would hypothesize the process is a reversal of the previously studied Wallerian degeneration (discussed above). In contrast to the degeneration process, during the rejuvenation process we hypothesize that the surviving visual neurons increase their myelination and synaptic and dendritic complexity.

An interesting pattern in the Argus II group data (albeit not the case studies) is the lateralization of significant changes in cortical thickness to the left hemisphere. Lateralization of significant reorganization is not unusual in group studies in the blind. For example, the right cuneus cortex and occipital gyrus were found to respond to an auditory-spatial task (Collignon et al. 2011), and the right occipital cortex was activated by odor identification (Renier et al. 2013) in the blind. Whereas, the blind displayed left visual cortex activation during sound localization (Voss et al. 2008; Renier et al. 2013), and left ventral lateral occipital complex utilization in a verbal control task (Bedny et al. 2012). Unfortunately, studies with blind participants are both small (10 to 20 participants) and noisy; therefore, lateralization of results could be due to brain dynamics (such as with language) or could be due to the limitations of a given data set (Lazzouni and Lepore 2014). Future studies that have larger participant groups or within-participant baselines (such as before and after blindness onset) are necessary to better evaluate these alternative hypotheses.

Furthermore, the results in this paper are also limited by the cohort size; nevertheless, they contribute to the broader literature on cortical adaptation to vision restoration in the late blind. It is likely that the formulation of the overarching theory of rejuvenation dynamics will require multiple studies of neuroimaging with a variety of visual restoration participants. This paper furthers the research toward that goal.

A remaining question is the relationship between this cortical rejuvenation and the behavioral outcomes with artificial vision. In other words, does the cortical rejuvenation impact the level of functionality of individuals with the Argus II device? Our behavioral measure of visual shape matching did not significantly correlate with visual cortical thickness in the Argus II participants. However, this nuanced function–structure relationship likely either requires a larger variety of behavioral tasks (as Argus II participants functionality can vary depending on task) or a larger cohort to be fully elucidated.

We next discuss the 2 regions that had significant rejuvenation in our group level analysis, the lateral occipital cortex and the cuneus cortex, and their potential relationship to visual function in the visually restored.

The lateral occipital cortex and vision restoration

The Argus II participants’ thicker left lateral occipital cortex relative to late blind controls (and similar to sighted controls) is likely due to the restored visual function from the artificial vision. While the comparison between participant groups is not causal (ie a pre- and post-measure), it does indicate a significant difference relative to blind controls with the same retinal disease (RP), comparable ages and a similar level of natural vision loss.

Lateral occipital cortex (Fig. 2A) in the Desikan–Killiany atlas covers multiple sulci and gyri within the occipital lobe [description in the appendix of Klein and Tourville (2012)] including portions of the transverse occipital sulcus and the lateral occipital sulcus. The lateral occipital cortex is the highest level in the visual processing hierarchy within the occipital lobe in the Desikan–Killiany atlas, in particular, the visual regions progress from the pericalcarine cortex (including most of the primary visual cortex) up to the lateral occipital cortex (including mostly V2 and V3). The lateral occipital cortex has been shown (using neuroimaging techniques) to process objects’ 3D shape (Doniger et al. 2000; Grill-Spector et al. 2001). Furthermore, individuals with lesions in the lateral occipital cortex have been found to have deficits in the recognition of object shape (Moscovitch et al. 1997; Ptak et al. 2014). In addition, transcranial magnetic stimulation (TMS) that suppresses the activity of the lateral occipital cortex can generate a deficit in the object recognition process (Stewart et al. 2001). Recent TMS research points to a feedback of perceived object size information (in contrast to earlier measures of retinal size) from the lateral occipital cortex to early visual cortex (Zeng et al. 2020). The lateral occipital cortex has also been shown to process multisensory information, in particular, integrating shape information between tactile, visual, and auditory modalities (Beauchamp 2005; Amedi et al. 2007).

Argus II participants have been shown to be able to recognize basic objects (Luo et al. 2014) and integrate shape across the senses (Stiles et al. 2022) with artificial vision. Therefore, the reinvigoration of the lateral occipital cortex could be an important part of rehabilitating these basic visual functions with artificial vision. Furthermore, the lateral occipital cortex’s direct feedback connections to earlier visual cortices [shown in previous TMS experiments (Zeng et al. 2020)] may have a critical role in the early processing of visual and multisensory information in the visually restored brain. In general, the hypothesis that a thickened left lateral occipital cortex could facilitate the rehabilitation of Argus II participants is consistent with the results in this paper and in the literature.

The cuneus cortex and vision restoration

The Argus II participants also had a significantly thicker left cuneus cortex relative to blind controls. The cuneus cortex (Fig. 2A) in the Desikan–Killiany atlas covers multiple sulci and gyri within the occipital lobe (description in appendix of Klein and Tourville (2012)] including portions of the parieto-occipital sulcus, and the posterior limit and dorsomedial margin of the calcarine sulcus. The parieto-occipital sulcus within the cuneus cortex has been shown (using neuroimaging techniques) to process mental navigation and egocentric spatial tasks in humans and nonhuman primates (Ino et al. 2002; Pitzalis et al. 2021). A lesion study in nonhuman primates connects the parieto-occipital sulcus with visuomotor control, in particular lesions generated deficits in reach and grasp tasks (Battaglini et al. 2002; Galletti et al. 2003). Therefore, whereas the lateral occipital cortex is in the ventral stream and processes the identity of objects, the cuneus cortex is located in the dorsal stream and is attuned to the location and orientation of the body relative to objects. The Argus II participants are trained to navigate with their devices and have shown the utility of visual information in mobility tasks (Jeganathan et al. 2022; Flourence et al. 2023; Sadeghi et al. 2024). Therefore, the thickening of the cuneus cortex observed in this study is consistent with the functionality measured in Argus II participants.

Conclusions

Overall, the Argus II participants were shown to have a thicker cortex relative to the blind controls in 2 higher visual regions (the cuneus cortex and the lateral occipital cortex). A correlation analysis showed that longer using Argus II participants had on average a thicker visual cortex and 2 pre- and post-implantation case studies fortified this trend. Therefore, these results support the theory that the visual cortex can be structurally rejuvenated with vision restoration following decades of late blindness and visual cortical atrophy.

Author contributions

Noelle Stiles (Conceptualization, Funding acquisition, Investigation, Methodology, Project administration, Visualization, Writing—original draft, Writing—review & editing), Jeiran Choupan (Investigation, Methodology, Writing—review & editing), Hossein Ameri (Investigation, Writing—review & editing), Vivek Patel (Project administration, Supervision, Writing—review & editing), Yonggang Shi (Funding acquisition, Methodology, Project administration, Supervision, Writing—review & editing).

Funding

National Institutes of Health, National Eye Institute (1U01EY025864-01), National Institutes of Health, the BRAIN Initiative (5K99EY031987-02), Arnold O. Beckman Postdoctoral Scholars Fellowship Program, USC Roski Eye Institute, Philanthropic Educational Organization Scholar Award Program.

Conflict of interest statement

None declared.

Data availability

Neuroimaging data will be made available through the Connectome Coordination Facility (https://www.humanconnectome.org/) as a part of the Human Connectome for Low Vision Project funded by the National Institutes of Health.