simulating prosthetic vision: ii. measuring functional ... · pdf filesimulating prosthetic...

Vision Research 49 (2009) 2329–2343

Contents lists available at ScienceDirect

Vision Research

journal homepage: www.elsevier .com/locate /v isres

Minireview

Simulating prosthetic vision: II. Measuring functional capacity

Spencer C. Chen a,b,*, Gregg J. Suaning a, John W. Morley b,c, Nigel H. Lovell a

a Graduate School of Biomedical Engineering, University of New South Wales, Kensington 2052, Australiab School of Medicine, University of Western Sydney, Penrith, NSW 1797, Australiac School of Medical Sciences, University of New South Wales, Kensington 2052, Australia

a r t i c l e i n f o a b s t r a c t

Article history:Received 16 July 2008Received in revised form 9 July 2009

Keywords:Prosthetic visionSimulated prosthetic visionPsychophysicsFunctional visionVisual acuity

0042-6989/$ - see front matter � 2009 Elsevier Ltd. Adoi:10.1016/j.visres.2009.07.003

* Corresponding author. Address: Graduate SchooUniversity of New South Wales, Kensington 2052, Aust

E-mail address: [email protected] (S.C. Chen).

Investigators of microelectronic visual prosthesis devices have found that some aspects of vision can berestored in the form of spots of light in the visual field, so-called ‘‘phosphenes”, from which more rich andcomplex scenes may be composed. However, questions still surround the capabilities of how such a formof vision can allow its recipients to ‘‘see” and to carry out everyday activities. Through simulations ofprosthetic vision, researchers can experience first-hand many performance and behavioral aspects ofprosthetic vision, and studies conducted on a larger population can inform the performance and behav-ioral preferences in general and in individual cases. This review examines the findings from the variousinvestigations of the functional capacity of prosthetic vision conducted through simulations, especiallyon the topics of letter acuity, reading, navigation, learning and visual scanning adaptation. Central tothe review, letter acuity is posited as a reference measurement so that results and performance trendsacross the various simulation models and functional assessment tasks can be more readily comparedand generalized. Future directions for simulation based research are discussed with respect to designinga functional visual prosthesis, improving functional vision in near-term low-phosphene-count devices,and pursuing image processing strategies to impart the most comprehensible prosthetic vision.

� 2009 Elsevier Ltd. All rights reserved.

1. Introduction

The human implantation of a visual prosthesis device by Brind-ley and Lewin in 1968 saw the successful elicitation of artificialvisual perception, so-called ‘‘phosphenes”, described as spots oflight ‘‘the size of a grain of sago at arm’s length” or ‘‘like a star inthe sky”. Similar visual sensations have been confirmed in subse-quent human trials of past and modern visual prosthesis devices(for example, Brindley & Rushton, 1974; Dobelle 2000; Dobelle,Mladejovsky, & Evans, 1976; Dobelle, Mladejovsky, & Girvin,1974; Humayun et al., 2003; Richard, Hornig, Keseru, & Feucht,2007; Richard et al. 2005; Rushton & Brindley, 1978; Veraartet al., 1998; Zrenner et al., 2006, 2007). Elicitable in various sizes,luminance intensity and shapes, this rudimentary form of restoredvisual perception is considered to be the fundamental buildingblock for creating visual scenes filled with rich and complex pat-terns described as ‘‘prosthetic vision”.

Nevertheless, vision is more than the mere perception of spotsof light in various sizes, luminance and shapes. A visual prosthesismay be able to elicit visual percepts from electrical stimulation at

ll rights reserved.

l of Biomedical Engineering,ralia. Fax: +61 (2) 9663 2108.

the retina, optic nerve or at the primary visual cortex, but the prob-lem of visual comprehension lies in how the implant recipientinterprets such information.

Using an idealized simulation of prosthetic vision as an example(Fig. 1), amongst many things, the most noticeable feature is thediscreteness of the phosphenes; there are large gaps with no visualinformation in between the phosphenes, as opposed to a perceptu-ally continuous visual image in normal vision. Consequently, sepa-ration of groups of phosphenes portraying one object over groupsof phosphenes portraying another is required. In addition, currenttechnology limits the elicitation of one or only a handful of phos-phenes at any one instance, and even if all phosphene can besimultaneously perceived, the limited number of stimulating sitesresults in very low-resolution vision, the phosphenes are unevenlydistributed over a reduced field of view, and the dynamic range ofphosphenes limits the contrast presentable.

Many questions still surround how to best utilize such a formof vision to present an understandable visual perception. An inti-mately related problem is the training and rehabilitation rou-tines to assist recipients in gaining the maximum benefit froma visual prosthesis device. Past attempts at developing aids forthe blind by converting visual into auditory or tactile form hasresulted in low acceptance due to the difficulty in interpretingthe converted signals (Hungenahally, 1995). Therefore, it iscrucial that prosthetic vision be approached from the functional

http://dx.doi.org/10.1016/j.visres.2009.07.003

mailto:[email protected]

http://www.sciencedirect.com/science/journal/00426989

http://www.elsevier.com/locate/visres

Fig. 1. Examples of phosphene vision. Left: Some text. Middle: A female face. Right: An office area.

2330 S.C. Chen et al. / Vision Research 49 (2009) 2329–2343

point of view, acquiring information from the behavioral,psychological, perceptual and cognitive domains, so that themost comprehensible visual presentation and most effectiverehabilitation routines can be formulated.

This review examines the methods by which researchers, usingsimulation studies of prosthetic vision, have assessed the level towhich recipients can see and perform tasks. Two shortcomings ofthe current research are identified. Firstly, there needs to be betterinteroperability in the results between research groups, the variousdifferent forms of assessment techniques and prosthetic visionsimulation models. Improved interoperability will allow research-ers to more clearly describe the trends in functional prostheticvision with respect to various contributing factors, rather than tosimply view each investigation as a separate case study. Secondly,more attention is required in designing assessment protocols inlight of the considerable learning trends observed in many SPVstudies. Investigators should pay special attention to behavioralmodification such as head and eye scanning movements. Finally,in the concluding section, simulation studies are related back tothe need for driving aspects of visual prosthesis design andimproving the functional capacity of the recipients implanted withnear-term low-resolution devices.

Table 1List of human trials with visual tasks examined.

Year Author + Notes Exercises

1962 Button and Putnam Light localizationNavigation

1972 Brindley et al. Visual Braille reading

1974 Dobelle et al. Visual Braille reading

2000 Dobelle et al. (1978 implants) Character recognitionLetter acuityMobility

2001 Dobelle et al. (Portugal implants) Object localizationNavigationDriving a car

2002 Veraart et al. Pattern recognitionOrientation discriminObject localizationObject discriminationHand-eye coordinatio

Humayun et al. Light detectionLight localizationMotion detectionObject detectionObject countingObject discriminationOrientation discriminMovement direction

2007 Second sight Movement direction(Argus II) Object localization

Zrenner et al. Letter acuityOrientation discrimin

2. Measuring functional capacity of prosthetic vision

Human performance can be assessed using a variety of psycho-physical techniques. Psychophysical assessments are tasks de-signed to identify particular characteristics of human (or animal)perception, cognition and performance by manipulating physicalsensory stimuli such as visual displays, sounds, and tactile texture,etc. These tasks are designed so a quantifiable performance mea-sure can be analyzed – such as reaction time, perception threshold,or success rate – so as to reveal the psychological and neural mech-anisms underlying the performance response.

In human trials, investigators have been interested in the abilityof the implant recipients in recognizing simple characters, objectsand patterns, and their ability to manipulate with their environ-ment through hand-eye coordination and navigation tasks (Table1). They demonstrate that given optimized conditions, implantrecipients were able to successfully identify and differentiate be-tween rudimentary patterns and objects, and conduct limitedand assisted navigation about a high contrast environment.

The description of the visual perception reported by the im-plant recipients provides investigators with an opportunity to fur-ther study the functional capacity of current devices and devices

References

Button and Putnam (1962)

Brindley and Rushton (1974)

Dobelle et al. (1976)

Dobelle (2000)

Unpublished

Delbeke et al. (2002), Veraart et al. (2003), Veraart,Duret, Brelen, and Delbeke (2004), Brelen et al. (2005),Duret et al. (2006)

ation

n (grasping)

Humayun et al. (2003, 2004), Weiland et al. (2003,2004), Yanai et al. (2007)

ationdetection

identification Ahuja et al. (2009), McMahon et al. (2009)

Wilke et al. (2009), Zrenner et al. (2009)ation

Table 2List of simulated prosthetic vision research.

Year Author Objectives References

1990 Cha et al. Letter acuity Cha et al. (1992a, 1992b, 1992c)ReadingEye trackingMobility

2001 Boyle et al. Object recognition Boyle, Maeder, and Boles (2001), Boyle,Maeder, and Boles (2002a, 2002b, 2003)

Dagnelie et al. Letter acuity Humayun (2001), Hayes et al. (2003), Thompson et al. (2003)Object recognitionReadingHand-eye coordinationFace recognition

2003 Hallum et al. Smooth pursuit Hallum et al. (2003, 2005)

Sommerhalder et al. Reading Sommerhalder et al. (2003, 2004), Perez Fornos et al. (2005)

2004 Chen et al. Letter acuity Chen et al. (2004, 2005a, 2005b, 2006, 2007)

Dowling et al. Mobility Dowling et al. (2004, 2005)

2005 Cai et al. Letter acuity Cai et al. (2005)

2006 Dagnelie et al. Reading Dagnelie et al. (2006a, 2006b)Hand-eye coordination

Fu et al. Reading Fu et al. (2006)

Vurro et al. Face recognition Vurro, Baselli, Orabona, and Sandini (2006)

2007 Chai et al. Character recognition Chai et al. (2007)

Dagnelie et al. Mobility Dagnelie et al. (2007)

2008 Chai et al. Phosphene mapping Chai et al. (2008)

Perez Fornos et al. Simple pointing Perez Fornos et al. (2008)Pattern identificationHand-eye coordination

Wang et al. Smooth pursuit Wang et al. (2008)

S.C. Chen et al. / Vision Research 49 (2009) 2329–2343 2331

under development using simulation studies of prosthetic vision.Table 2 lists the psychophysical tasks studied in the context ofsimulated prosthetic vision (SPV). These experiments sought toquantify the usability of prosthetic vision under SPV settings thatreflect current research devices. Such work was pioneered by Chaand coworkers, (1992a, 1992b, 1992c), but only until recently(year 2000 and beyond) has it attracted the interest of moreinvestigators owing to the rapid development of visual prosthesisdevices.

Inevitably, at this early stage of visual prosthesis development,there remains a gulf to be reconciled between the assumptionsmade in SPV setups and that of the experience of real implantrecipients. For example, phosphene shapes, color, sizes, phosphenefield eccentricity and regularity are simplified in simulations; andmuch is still to be learnt in order to characterize the interactive vi-sual effects to simultaneous multisite stimulation and temporalresolution of the elicited phosphenes. These were outlined in thefirst part of the two-part review on simulating prosthetic vision(Chen, Suaning, Morley, & Lovell, 2009a). Nevertheless, SPV studieshave been able to provide valuable input regarding the perfor-mance capacity of phosphene vision.

In the following subsections, the major findings from SPVresearch are summarized. Readers might also be interested inpast reviews on SPV research (Hallum, Dagnelie, Suaning, &Lovell, 2007; Hallum, Suaning, & Lovell, 2004a), and partialreviews on specific functional assessments (Dagnelie, Barnett,Humayun, & Thompson, 2006; Dagnelie et al., 2007, Perez Fornoset al., 2008).

2.1. Factors affecting measured performance

When interpreting the results from different laboratories, it isimportant to be aware of the various factors that can affect the

measured performance. The main factors are the configuration ofthe SPV and the task design.

In a previous review, various aspects of the SPV implementa-tions were summarized and discussed (Chen et al., 2009a). Briefly,a SPV model consists of the visual appearance of the phosphenes(solid circles vs. Gaussians), the layout of the phosphenes in the vi-sual field (the phosphene count, spacing, regularity, field of viewand eccentricity), the processing that converts the original imageinto phosphenes (filtering scheme and the number of gray levels)and how the spatiotemporal dynamics of the phosphene field ispresented on a computer screen (static or dynamic presentation,and spatiotemporal interactions between phosphenes).

For example, increase in phosphene to background contrastslightly improved the performance in paragraph reading (Dagne-lie, Barnett, et al., 2006) and face recognition (Thompson, Barnett,Humayan, & Dagnelie, 2003), and increase in the number of graylevels led to improved face recognition performance, but not par-agraph reading; a dynamically updated phosphene vision basedon the gaze position of the subject improved reading performanceover an otherwise static display (Perez Fornos et al., 2005). Inves-tigators vary or control some of these parameters as they studythe affects of these parameters on the overall performance, someof which are summarized below. However, as the SPV configura-tion reported from each laboratory can be quite disparate, it isnot possible to neither compare nor discuss all the factors indetail.

A variety of psychophysical tasks have been used to assess thefunctional capacity of SPV (Table 2). In terms of methodology, thereare hardly any two tasks that have been repeated across differentlaboratories. For example, the assessment of letter acuity con-ducted in different laboratories may involve testing on differentsets of test items (Landolt C or Snellen E, etc.) and different meth-ods of scoring. Similarly, the amount of practice can have a bearing


on the performance due to adaptation and learning. In addition, themethod through which scanning is provided – whether it is hand-,head-, or eye-facilitated – may have a significant impact on theperformance outcome. Topics relating to learning and visual scan-ning are covered in more detail in a latter section.

2.2. Reading

Reading, by far, is the most studied psychophysical task under SPV(Cha, Horch, & Normann, 1992a; Cha, Horch, Normann, & Boman,1992b; Hayes et al., 2003; Sommerhalder et al., 2003; Sommerhalderet al., 2004, Perez Fornos et al., 2005, Dagnelie, Barnett, et al., 2006; Fu,Cai, Zhang, Hu, & Zhang, 2006). There are two main measures of read-ing performance – reading accuracy and reading speed.

Reading accuracy measures the percentage of the words readcorrectly. Near 100% accuracy is required for optimal reading speedand 85% accuracy is required for a (‘‘good”) level of text compre-hension such that subjects could grasp the main issue in the textbut not all the details (Sommerhalder et al., 2004).

Reading speed is often reported in words per minute (wpm).Normal reading can easily reach more than 200 wpm (Legge, Pelli,Rubin, & Schleske, 1985). Optimal reading requires at least a visualwindow of four characters (Legge et al., 1985). A critical samplingrate of at least three to four phosphenes per axis per character isrequired for individual letter recognition (Akutsu, Bedell, & Patel,2000; Cha et al., 1992a), and up to five to six phosphenes per char-acter for the optimum reading speed (Cha et al., 1992b; Dagnelie,Barnett, et al., 2006; Fu et al., 2006; Legge et al., 1985; Solomon& Pelli, 1994). Insufficient resolution rapidly reduced the readingspeed. For multi-line (paragraph) reading, a vertical window of atleast two lines of visible text is preferred (Sommerhalder et al.,2004, Perez Fornos et al., 2005).

Eccentric reading has been studied especially by Sommerhalderand co-workers (Sommerhalder et al., 2003, 2004, Perez Fornoset al., 2005). They showed that reading speed declines with increasingeccentricity. The decline is associated in part with the loss of visualresolution at eccentric retinal locations and the ‘‘crowding” effect.

The reading characteristics of non-Roman alphabets are differ-ent and require separate investigation. For example, Chai and co-workers (2007) conducted SPV experiments on reading Chinesecharacters. Chinese characters are designed to each occupy asquare block, but the composition complexity (spatial frequencycontent) varies dramatically from character to character. Usingan optimized presentation of these characters, Chai and co-work-ers found that 12 phosphenes per axis (as opposed to four for theRoman alphabet) is required to achieve close to 100% individualcharacter recognition.

2.3. Mobility

Several investigators have studied mobility in SPV (Cha et al.,1992c; Dowling, Maeder, & Boles, 2004, Dowling et al ., 2005,Dagnelie et al., 2007). The general setup involves a controlled cor-ridor or room with fixed-position ground level and/or overhangingobstacles. The subject has to navigate from one location to anotherwithin the controlled environment, viewing through a mobilizedform of the SPV apparatus. In one study, a computer maze wasset up such that subjects navigated virtually using a game control-ler (Dagnelie et al., 2007). Performance was assessed on the timetaken to finish the course, the number of obstacle contacts, andother error measures specific to the course design.

Cha et al., 1992c found that the visual field size and number ofpixels had high correlations with navigation performance. Theyestimated that at least 30� of the visual field is required for a goodlevel of navigation ability. (Note: Cha and co-workers assessed thisby minifying a larger field of view to a 1.7� � 1.7� phosphene field.)

Information from the visual field up to 58� eccentricity is likely tomake significant contributions to safe mobility and orientation(Lovie-Kitchin, Mainstone, Robinson, & Brown, 1990).

Overall, it is difficult to relate the performance measures fromone laboratory to another in mobility assessment due to the widevariety of assessment methodology, courses and maze settings,and obstacle difficulty.

2.4. Other tasks

As listed in Table 2, normally-sighted subjects under SPV havebeen assessed in object recognition (Hayes et al., 2003), patternmatching (Perez Fornos et al., 2008), face recognition (Thompsonet al., 2003), smooth pursuit (Hallum, Suaning, Taubman, & Lovell,2005; Hallum, Taubman, Suaning, Morley, & Lovell, 2003; Wang,Yang, & Dagnelie, 2008), simple pointing (Chai et al., 2008; PerezFornos et al ., 2008) and hand-eye coordination tasks (Dagnelie,Walter, & Liancheng, 2006; Hayes et al., 2003, Perez Fornoset al., 2008).

Dagnelie and co-workers (Dagnelie, Walter, et al., 2006; Hayeset al., 2003) have been focusing on SPV configurations similar tothe clinical trial devices prepared by Second Sight (Ahuja et al.,2009; Humayun et al., 2003; McMahon et al., 2009), i.e., 4 � 4and 6 � 10 electrodes. They showed that some degree of letter acu-ity, pattern/object identification, and visually coordinated handmanipulation is possible, despite only a handful of phosphenesbeing available to convey the visual information of the entirescene.

Face recognition requires a larger phosphene lattice. It is gener-ally agreed that spatial frequencies in the range 8–16 cycles perface width are critical for face recognition (Costen, Parker, & Craw,1994; Gold, Bennett, & Sekuler, 1999; Nasanen, 1999). This meansto render a face requires a 16 � 16 to 32 � 32 phosphene lattice.Thompson and co-workers (2003) investigated this in SPV, andfound that 11.1 phosphenes per axis may be sufficient to recognizefaces.

Smooth pursuit experiments (Hallum et al., 2003, 2005; Wanget al., 2008) demonstrate the motion sensitivity in SPV. Comparedto normal vision, the detection of the onset of movement is re-tarded, and tracking is less accurate using prosthetic vision, andeven less accurate when an eccentric retinal locus is adopted fortracking.

Perez Fornos and co-workers (2008) investigated the accuracyin pointing and matching patterns on a 5 � 4 set of patterned woo-den chips over a range of visual field sizes and phosphene resolu-tions. It was found that reduced visual field increased the searchtime and reduced the pointing and pattern matching performanceat low phosphene counts. They found that for pointing, optimumperformance was attained at a phosphene resolution of 1.5 phos-phenes per degree, and for pattern matching, optimum perfor-mance requires a phosphene resolution of two phosphenes perdegree.

Mainly due to the specific designs of these tasks, not many lab-oratories have published comparable studies. There are concernsover the validity of generalized trends based on results obtainedon specific SPV configurations and designed tasks.

2.5. The need for a united approach

The main drawback of the various reported study methodolo-gies is their insufficiency as controlled measurements that can beused repeatedly in different laboratories with consistent, quantita-tive and comparable results. Take reading for example: althoughfonts, character, and line spacing can be controlled, there is a vari-ety of reading materials to choose from for assessment; each inves-tigator tends to prefer their own formulation and level of difficulty.


Further, investigators configure SPV with respect to their owndevelopmental devices, to the limitations of their simulation hard-ware, or merely to their own preferences. Given this current stateof SPV research, it is difficult from currently available data to buildinformative psychophysical models of visual prosthesis recipients.

A simple set of reference measurements would provide a meansof comparing across the spectrum of SPV studies and human clin-ical trial data. This is not to say that only one or a few measure-ment indices are required for characterizing the usability ofprosthetic vision for all types of daily activity; nonetheless, onecan continue to invent new specialized tasks to assess the usabilityof prosthetic vision under different scenarios, but being special-ized, they preclude a comparable study from being easily set up.On the other hand, if a set of reference assessments can be assem-bled that demonstrate a high degree of correlation to various visu-ally guided tasks, then performing and reporting thesemeasurements can allow generalization to other SPV configura-tions and visual prosthesis designs. A concerted approach usingreference assessment techniques will better inform prosthetic vi-sion researchers, for example, on several design aspects of a visualprosthesis, such as, but not limited to, the number of electrodes,their spacing, visual field coverage and image processing, so thatone design can be compared with another in order to find theoptimum.

This review proposes letter acuity for consideration as one ref-erence assessment for evaluating the functional capacity of pros-thetic vision. The performance in various psychophysical tasksassessed in SPV studies in relation to letter acuity as the referencemeasure will be discussed. Other basic reference assessmentsshould also be designed so as to reflect other major factors contrib-uting to functional prosthetic vision, such as incorporating contrastsensitivity assessment to letter acuity tests (e.g., Pelli-Robson andRegan charts), and assessing the effective field of view providedby the phosphene field.

3. Letter acuity as a reference measure for prosthetic vision

Visual acuity assessed through lettered charts, i.e. ‘‘letter acu-ity” (LA), has always been the foremost clinical screening test ofhuman visual function. It is a good candidate as a reference mea-surement for SPV investigations. Various quality of life surveysconducted on sufferers of retinitis pigmentosa (Geruschat, Turano,& Stahl, 1998; Szlyk et al., 1997, 2001), macular degeneration(Hazel, Petre, Armstrong, Benson, & Frost, 2000; Mangione,Gutierrez, Lowe, Orav, & Seddon, 1999; Mcclure, Hart, Jackson,Stevenson, & Chakravarthy, 2000), diabetes retinopathy (Klein,Moss, Klein, Gutierrez, & Mangione, 2001), patients scheduled fora corneal graft (Boisjoly et al., 1999), and in cohorts involvingpatients from a variety of low vision symptoms (Carta et al.,1998; Haymes, Johnston, & Heyes, 2002; Mangione et al., 2001)have repeatedly demonstrated correlations between LA and manydaily tasks. For the visually impaired, diminished LA has been asso-ciated with decreased performance of activities of daily living,poorer cognitive abilities, increased risk of falls, and ultimatelypoorer health-related quality of life.

LA is a clinical test that assesses the observer’s ability to recog-nize simple test items (optotypes). Optotypes are usually based onthe English alphabet such that each contains the integration of sev-eral visual features. Yet they are designed so that the smallest sizeat which they can be correctly identified (the LA threshold) esti-mates the finest sinusoidal components that can be perceived bythe observer. Given a visual scene can be seen as a linear combina-tion of elemental sinusoidal components of different (orthogonal)periodic occurrences at different orientations, therefore, establish-ing the LA threshold also has important implications for investiga-

tors working on image processing. Both clinically and analytically,LA is an appropriate, validated, indicative and fundamental ratingfor functional vision under different phosphene configurations.

LA has been used both in human trials of prototype visual pros-thesis devices (Dobelle, 2000) and in SPV experiments (Cai, Fu,Zhang, Hu, & Liang, 2005; Cha et al., 1992a; Chen, Chen, Hallum,Suaning, & Lovell 2006; Chen, Hallum, Lovell, & Suaning 2005b;Chen, Hallum, Suaning, & Lovell 2007; Chen, Lovell, & Suaning2004; Hayes et al., 2003). In particular, Chen and co-workers(2004, 2005b) described a LA study paradigm that achieved statis-tically significant comparisons between hexagonal and squaremaps of phosphenes, and across a range of low-pass cut-off fre-quencies for image processing. This LA study paradigm can possi-bly be the basis of a more pervasive reference assessment for thefunctional capacity of prosthetic vision.

3.1. Letter acuity, phosphene lattice and image processing

Traditionally, SPV studies are based on regular, square latticeswith the phosphenes aligned in the horizontal and the verticaldirections. Mathematical calculation shows that the hexagonal lat-tice is 13.4% higher in electrode density over a square lattice of thesame minimum phosphene center-to-center spacing (PS); that is,given the same number of sampling points (phosphenes), hexago-nal sampling provides a higher sampling resolution (therefore LA)over the square lattice. Therefore, it is expected that phosphenesrendered on a hexagonal lattice would afford higher LA scorescompared to the traditional square lattice.

The other aspect of SPV investigated by Chen and co-workers(2004, 2005b) is whether changes in the low-pass filtering cut-off can be reflected in LA measurements. Averaging filters with acircular aperture (circular mean filters, CMFs) were examined. ACMF with a large radius has a low low-pass cut-off frequency, thusit would attenuate high frequency components contained in thephosphenized image, removing the important spatial details thatare required for the correct identification of the test symbols(Fig. 2). Conversely, a CMF with a small radius would lend itselfto a high low-pass cut-off frequency, retaining the frequency com-ponents that introduce aliasing artifacts to the low-resolutionphosphene field, thereby reducing the recognizability of the objectin view. The hypothesis is that towards the extreme values of theradius (i.e. low-pass cut-off frequency), the LA performance willdecline for the reasons suggested, and an optimally performing ra-dius setting may be found in between the extremes.

Both of the described problems involve only a small perturba-tion to the SPV configuration. If the measurement variance usingthe LA metric is large, then the statistical performance differencesdue to these small perturbations would be hard to determine. If theLA metric has the ability to demonstrate these hypotheses quanti-tatively with statistical significance, then it has the potential to bemore widely applied to compare the performance of other adjust-ments to SPV configurations, such as phosphene density, appear-ance and other image processing routines.

Chen and co-workers (2004, 2005b) administered SPV to nor-mally-sighted subjects through virtual-reality simulation pre-sented on a head-mounted display. LA was tested with theLandolt C optotype using the method of constant stimuli and esti-mated by fitting a psychometric function to the percentage perfor-mance versus optotype size (in logarithm of Minimum AngularResolution (logMAR)). This assessment routine ensured an accurateand consistent estimate of the LA thresholds.

It was found that the LA performance on the hexagonal phos-phene lattice has a small (0.05 logMAR with their SPV configura-tion) but statistically significant (p < 0.0001) advantage over asquare phosphene lattice with a matched minimum PS. As forthe filter settings, the LA performance across CMF radius values

Fig. 4. Relationship between the estimated letter acuity (LA) thresholds and thephosphene center-to-center spacing (PS) compiled from the vision prosthesisliterature. Black dots indicate data points from simulated prosthetic vision work.The star represents the only available LA data from human trials of actual implant.The shaded area indicates the approximate 95% confidence interval of the linear fit(solid gray line). The theoretical LA threshold with respect to PS is represented bythe dotted gray line. Note: Increasing logMAR is increasing physical size andworsening LA.

Fig. 2. The effect of image filtering using the circular mean filter with increasing radius (left to right). The Landolt C test item and the orientation of the gap becomeincreasingly unrecognizable due to the lack of detail, both in the non-phosphenized view (top row) and the phosphenized view (bottom row).

Fig. 3. Data of the letter acuity (LA) estimates plotted against circular mean filterswith a range of radius values (expressed in terms of phosphene center-to-centerseparation (PS)) from Chen et al. (2005b) replotted with quadratic fitting. The R2

values of the fittings are displayed.

1 0.39 logMAR is about 2.50 of visual angle or 12 lm on the retina.


exhibited the expected ‘‘U” shaped trend. Quadratic models werefitted to the results with statistically significant coefficients(p < 0.05) as shown in Fig. 3. These results demonstrated LA as aconsistent and sensitive measure that is able to follow fine-grainchanges to the SPV configuration, enough to allow modeling theLA performance over a multi-parametric SPV configuration space.

Subsequently, Cai and co-workers (2005) also used LA (LandoltC) as a comparison metric to study the performance impact ofirregular phosphene maps. They adopted a similar experimentalparadigm to that described by Chen and co-workers (2004,2005b). They showed that the perceptual quality of prosthetic vi-sion decreased with increasingly irregular phosphene maps, espe-cially if the visual locations of the phosphenes were inaccuratelyrecorded in the phosphene processing software. Unfortunately,statistical analysis was not included in their report.

3.2. Letter acuity and phosphene spacing

In other SPV studies, it is not surprising to find that the mostinfluential SPV parameter on LA thresholds is the PS. This was wellestablished early by Cha et al. (1992a). Results from Cha and co-workers’ study and other investigations have been compiled inFig. 4. As shown, a linear relationship is evident between PS and

LA. The fitted gradient is 1.0; this means that for every unit reduc-tion in PS, the LA threshold will also reduce by one unit, providingthe phosphene size can also be reduced correspondingly so as toprevent neighboring phosphenes bridging together and blurringthe fine details of the underlying image. This linear model servesas a guide to estimate the LA threshold for a given electrode arraydesign based on the electrode spacing and the expected visuotopicmapping of the target tissue.

The linear model fitted to the data predicts that 0.0 logMAR(normal vision) can be attained with 0.39 logMAR PS1 (p < 0.05),indicating that equivalent-to-normal-vision LA can be attained with-out a phosphene lattice matching the resolution of the human pho-toreceptor array (i.e., 0.0 logMAR). In fact, all studies observed LAperformance better than the theoretical (sampling theory) estimateof the phosphene field on which the subjects were assessed (dottedgray line).

Fig. 5. Reading speed in words per minute (wpm) from various simulated prosthetic vision investigations in the literature plotted against the phosphene center-to-centerspacing (PS) of the phosphene lattice from which they were obtained. Note: Different methods and reading materials were adopted in each laboratory.


This observation is particularly intriguing and requires furtherverification in visual prosthesis recipients. However, it may indi-cate the manifestation of mechanisms other than those based onconventional sampling theory in phosphene vision. For example,as was previously discussed, the phosphene lattice and filter set-tings both have an impact on the LA (Landolt C) performance (Chenet al., 2005b) – the data point representing these results is posi-tioned slightly lower (i.e. better) than the average trend (thin grayline, Fig. 4). The study performed by Cai and co-workers (2005,Landolt C) using an optimized filter setting that was based on Chenand co-workers’ findings was also lower than the average trend.Also, at the same PS, the number of electrodes (i.e. field of view)can affect LA performance; Hayes and co-workers (2003) demon-strated 4 � 4 and 10 � 6 phosphene arrays both with 2.08 logMAR(2� visual angle) PS, but the LA (Tumbling E) threshold for the 4 � 4array was higher (i.e. worse) than that of the 10 � 6 array.

Compared to SPV studies, Dobelle (2000) is the only investigatorwho has reported LA (Snellen letters, Tumbling E, HOTV, Landolt Cand Lea figures) estimates from human trials of a visual prosthesis.He reported an LA threshold approximately 1.78 logMAR fromelectrodes spaced approximately 1.33� visual angle or 1.90 log-MAR2. Dobelle’s result fits well within the predicted trend fromSPV studies in Fig. 4. Hopefully in the near future, results frommore human trials will further confirm this observed LA trend fromSPV studies.

3.3. Letter acuity and reading

While PS is chiefly a physical measurement, one of the manyparameters that an investigator may choose to vary or control inan assessment of functional performance, LA is an index of visualfunctional, whose value takes into account changes in PS as wellas many other factors that affect the quality of prosthetic vision.Therefore, besides an index for comparing performance of differentSPV configurations it is postulated that LA can be further general-ized into a more pervasive indicator of the functional capacity ofprosthetic vision. Once the main effect of LA is factored into theanalysis, performance differences due to other contributing factors

2 The electrode spacing of Dobelle’s cortical implant was 8 mm. Visual angle wasestimated using an estimate of 6 mm of cortex per degree of visual angle at the fovea(Cha et al., 1992a).

can be more clearly identified. In the following subsections, perfor-mance in reading, navigation and other tasks are compared to thePS due to the lack of direct LA measurements from the relevantstudies. Given the apparent relationship between LA and PS, infer-ences can be made of these performances to LA.

There is little doubt that LA is an important predictor for effec-tive reading (Fig. 6A). In a study involving low vision patients, theoptimal print size for reading was best predicted from LA thresholdestimates, and the word reading acuity was the best predictor forpeak reading speed (Bullimore & Bailey, 1995).

For SPV studies, the reading speeds reported by various investi-gators were compiled and plotted against the PS of the phosphenelattice from which they were obtained (Fig. 5). In general, the read-ing speed decreased as the PS increased; but no simple mathemat-ical model between PS and the reading speed can be inferred fromthe data; this is because reading speed is quite sensitive to theassessment methodology each investigator has chosen. For exam-ple, while both assessed using paragraph (multi-line) reading incentral vision, reading speeds reported by Dagnelie and co-workers(2006a) were substantially slower than the reported by Sommerh-alder and co-workers (2004, central).

In the case of Cha and co-workers (1992b) and Fu and co-work-ers (2006), their primary interest was the effect of PS on the read-ing performance; the reported data points from these twoinvestigators follow approximately linear trends with respect toPS measured in logMAR. However, it should be noted that reduc-tions in these reported reading speeds were influenced by bothan increase in PS and a reduction in the number of phosphenesto maintain a fixed field of view. If the phosphene field was infi-nitely large such that it did not play a role in restricting the readingwindow, the reading speed might have gradually approached zeroas the PS approached infinity; i.e., LA in logMAR is negatively cor-related to the logarithm of reading speed (Mcmahon, Hansen, &Viana, 1991).

Future studies should be conducted so that the reading acuityand the reading speed can be simultaneously measured, and theconfiguration parameters be varied in small steps to establishclearer models of human performance. This can be done usingstandard reading charts such as the MNREAD (e.g. Fu et al., 2006)or Radner Reading test charts.

In addition to the LA threshold estimate, the time taken for thesubject to identify a test item can be used as a rough estimate ofthe reading speed. For example, say it takes 1 s on average to iden-

Fig. 6. A phosphenized view of some text (row A), an office area (row B) and a female face from the Psychological Image Collection at Stirling University (row C) withincreasing spatial detail (letter acuity). From left to right, each successive image is composed of phosphenes on average at half the phosphene center-to-center spacing of theprevious image, i.e. twice the spatial frequency content. The full-resolution image is illustrated on the far right.

Fig. 7. A subset of the results from Thompson et al. (2003) demonstrating the lossof face recognition performance under both high and low contrast settings as thephosphene center-to-center spacing (PS) increases while all other simulationparameters were unchanged. Error bars indicated standard error of the mean.


tify one test item, for sentences with an average of five letters perword, this means an estimated reading speed of 60 letters per min-ute or 12 wpm. Practically, if subjects are able to recognize fourletters simultaneously, providing there is sufficient phospheneresolution and reading window, the new estimate would be 240letters per minute or 48 wpm.

3.4. Letter acuity and mobility

Though mobility studies are not to be viewed purely from a LAstandpoint, increasing spatial frequency information from theunderlying image will certainly improve navigation performanceunder prosthetic vision in real-life environments (e.g. Fig. 6B).There are several reports demonstrating statistically significantcorrelation between mobility performance and LA in low vision pa-tients (Brown, Brabyn, Welch, Haegerstrom-Portnoy, & Colenbran-der, 1986; Geruschat et al., 1998; Haymes, Guest, Heyes, &Johnston, 1996; Velikay-Parel et al., 2007). In these reports, a high-er correlation was found between mobility performance and LAwhen the navigation course involved a more complex environ-ment, increased assortment of obstacles and heavier human traffic(hospital foyer, small business area and shopping mall) than sim-pler and quieter environments (quiet corridor and residentialstreets).

From the handful of SPV studies, Cha and co-workers (1992c)and Dagnelie, Barnett and co-workers (2007) also showed thatthere might be a weak correlation between PS and mobility perfor-mance, inferring a weak correlation between LA and mobility per-formance. These assessments were performed in controlledenvironments with relatively large obstacles and low humantraffic.

More investigation is certainly required in the area of orienta-tion and mobility. While it would be hard to standardize the mobil-ity courses, investigators can improve upon the reporting detail oftheir SPV configurations, design more scientific rigorous assess-

ment protocols, attend to and account for the learning effect, andconduct the suggested LA reference assessment so as to assist theSPV community working in synergy to understand the various nav-igational difficulties pertaining to the utility of prosthetic vision.An example of work towards this end is being carried out by Veli-kay-Parel and co-workers (2007).

3.5. Letter acuity and other tasks

For each face or object, there would be a critical phosphene res-olution above which recognition would be trivial; below this crit-


ical point, recognition rate may need to be traded off with otherinfluencing factors, such as the extent of the visual field. Forexample, the critical resolution for face recognition is between 8and 16 cycles per face width (Costen et al., 1994; Gold et al.,1999; Nasanen, 1999). A subset of Thompson and co-workers(2003) assessments examined the effect of increasing PS on theperformance of face recognition. They studied these effects witha 16 � 16 square lattice, with phosphene size of 36.50 in diameterand PS of 36, 54 and 720. Their results are plotted in Fig. 7; thoughnot the sole contributor to performance in Thompson and co-work-ers’ report, the reduction in face recognition performance is evi-dent as PS increased (i.e. LA decreased). Studies with visuallyimpaired patients also demonstrated strong correlations betweenthe LA threshold of these patients and their face recognition perfor-mance (Bullimore, Bailey, & Wacker, 1991). An illustration of thebenefits of increasing spatial frequency information from theunderlying image to face recognition under prosthetic vision ispresented in Fig. 6C.

For object recognition, Hayes and co-workers (2003) asked theirsubjects to cut out a marked square on a piece of paper. Comparedto other lattices at 2� PS, only the 16 � 16 lattice with 450 PS provedto provide sufficient functional vision for this task.

In a recent study, Perez Fornos and co-workers (2008) assessedthe accuracy of locating LED light sources in space and matchingpatterns on wooden chips in SPV. Interestingly, a clear exponentialrelationship was found between the PS and the accuracy in bothassessments. It was found that optimum performance in their par-ticular setup can be attained at approximately a PS of 30 and 400 forthe pointing and pattern matching tasks respectively. Again, LA canbe inferred as a predictor to the performance in these tasks.

4. Learning and adapting to prosthetic vision

Irrespective of how functional prosthetic vision was measured,learning has been shown to contribute significantly to SPV task

Table 3List of simulated prosthetic vision investigations that showed significant contribution of l

References/Task Observed Learning

Cha et al. (1992c) During training, walking speed increased 5�Navigation Adaptation to the efficient use of head move

Chen et al. (2005a, 2005b, 2006, 2007) Learning across-session improved item identwith test items close to the threshold of idenLetter acuityIncreased head movement behavior observedIncreased scanning speed and adoption of ci

Dagnelie et al. (2006b) Considerable practice required for completinHand-eye coordination Adaptation of appropriate head scanning mo

Different strategies adopted by each subject,

Dagnelie et al. (2007) Subjects who have previous SPV experience nNavigation

Hallum et al. (2005) With practice, subjects able to fixate and tracsessions.Smooth pursuit

Perez Fornos et al. (2008) Time to complete both pointing and pattern msessions for pattern matching.Pointing

Pattern matchingHand-eye coordination

Sommerhalder et al. (2003) Word reading accuracy increased by 60% poiEccentric word reading Individual differences in reading strategy sus

Learned performance carried over to unpract

Sommerhalder et al. (2004), Perez Fornoset al. (2006)

Reading speed increased up to 5 � after pracinitially.

Eccentric paragraph reading Eye movements adapted to improve controlreflex.

Thompson et al. (2003) Response time reduced up to a third as a resFace recognition

performance. Learning can be quantified in two ways: improve-ment in the success rate and decrease in response time. The suc-cess rate will increase as the subjects become familiar with thesubtle details required to make informed judgments. This couldbe assisted by adapting to a specific method of viewing or scanningsuch that the required visual information can be extracted usingthe limited phosphene field. Subsequently, improvement in bothvisual perception and behavioral efficiency can lead to reduced re-sponse time.

Several investigators have published results profiling the learn-ing effect associated with prosthetic vision utility. For example,Chen and co-workers (2005a), Chen and co-workers (2005b) inan LA task demonstrated a learning effect of up to 0.2 logMARimprovement in LA threshold in some individuals. Similarly, learn-ing also had a dramatic influence on the reading accuracy of indi-vidual words (Sommerhalder et al., 2003), of paragraph text(Sommerhalder et al., 2004), accuracy in tracking a slow-movingtarget (Hallum et al., 2005), the time taken to count and placecheckers on a board (Dagnelie, Walter, et al., 2006), the time takento locate a light source and complete a pattern matching task (Per-ez Fornos et al., 2008), and walking speed in an obstacle course(Cha et al., 1992c) (see list in Table 3.)

The time course for learning differs from task to task, and islikely associated with the difficulty of the task and the aptitudefor learning in individual subjects. Note the learning time coursesfor the success rate and response time can also be different. Simpletasks such as pointing and pattern matching (Perez Fornos et al.,2008), LA assessment (Chen, Hallum, Lovell, & Suaning, 2005a;Chen et al., 2005b) or object tracking (Hallum et al., 2005) may take5–10 sessions to reach stabilized performance. Complex tasks suchas paragraph reading (Sommerhalder et al., 2004) or navigation(Cha et al., 1992c) may take 20 or 30 trials/sessions; in these tasks,much time would be spent familiarizing with the task as well aslearning to see in phosphene vision. In an eccentric reading taskunder SPV, Sommerhalder and co-workers (2003, 2004) noted that

earning.

.ments postulated.

ification by an average of 10% over ten sessions. Learning was twice the average ratetification., stabilized after about five sessions of practice.

rcular scanning strategy performed significantly better.

g tasks.vements with eye movement suppression in retinal-stabilized viewing mode.but given adequate practice, all subjects attained similar performance levels.

avigated through the course in a faster time and with less error than naïve subjects.

k the target up to twice the accuracy. Learning occurred mostly within the first four

atch tasks decreased with practice mainly over the first 10 sessions for point and 20

nts from practice. Response time almost halved after practice.pected.iced words, to the non-practiced eye and lasted over 6 months.

tice. Close to 100% reading accuracy from �10%

over horizontal scanning, vertical line shifts and suppression of fovea centering

ult of practice.

Fig. 8. Illustration of an exponential decay model of short term visual memory. Left: Nine consecutive frames of the same Landolt C test item displaced to different locationsof the visual field due to scanning. Scanning locus obtained from the unpublished data of the letter acuity experiment conducted by Chen and co-workers (2004, 2005b).Right: The result from applying the exponential decay model the nine phosphene frames in left, with the Landolt C from each frame centered with respect to each other.


the learning effect persisted for at least 6 months without practice.In the literature for perceptual learning tasks involving visualstimuli, it has been noted that the learned performance can be re-tained for 2–3 years without practice (Karni & Sagi, 1993; Sagi &Tanne, 1994).

Some reports are based on stabilized, learned results from theparticipating subjects (Cha et al., 1992a, 1992b; Chai et al., 2007;Fu et al., 2006) as opposed to results from ‘‘naïve”, inexperiencedsubjects to phosphene vision. In these reports, subjects may haveparticipated in former SPV studies or given a preliminary task tohelp familiarize them with the SPV environment, as it is believedthat learning in previous tasks, though different, may be trans-ferred to the new task. In a mobility task, Dagnelie and co-workers(2007) showed that subjects who had previously taken part intasks under SPV performed better than those new to SPV.

It is recommended that the issue of learning should always befactored into experimental designs and observations discussed togain insight towards patient rehabilitation.

4.1. The importance of visual scanning

Visual scanning in association with the use of prosthetic vision,whether in human trials of clinical prototypes or in SPV, is perhapsthe most cited behavioral adaptation. Scanning is a natural andspontaneous action to compensate for low visual resolution so asto improve perception (Bowers & Reid, 1997; Jan, 1991). Particu-larly for prosthetic vision, scanning of a scene is chiefly made nec-essary with the low resolution and discrete nature of thephosphene field. Scene scanning with a phosphene field is animportant skill to ‘‘fill-in” the lack of information in between thephosphenes.

In addition, though a static image of phosphenes may simplyappear like individual, isolated dots, when relative movement is in-duced via scanning, a more coherent structure of the sceneemerges. The existing literature on SPV tends to attribute this phe-nomenon to visual memory, mental imagery and spatiotemporalintegration (Cai et al., 2005; Chen et al., 2006; Dagnelie, Walter,et al., 2006; Hayes et al., 2003; Thompson et al., 2003), filling-inthe lack of visual information in the gaps between the discretephosphenes by accumulating the spatial visual information acrossframes of the visual presentation. As illustrated in Fig. 8, the orien-tation of the Landolt C gap is quite ambiguous judging by individ-ual frames. However, the gap can be clearly and unambiguouslyidentify when the nine successive frames are temporally integrated

using a simple exponential decay model for modeling temporalpersistence (Dixon and Di Lollo, 1994).

Scanning is also made necessary by the reduced visual fieldafforded by a phosphene field. In order to conduct visual search(Perez Fornos et al ., 2008) or to maintain spatial reference tovarious objects in the environment as in navigation (Cha et al.,1992c), the phosphene field has to be repositioned strategicallyin the visual space. The effective visual field is increased throughmechanisms similar to the aforementioned spatiotemporalintegration.

Under most of the present proposed visual prosthesis designs,eye movements are ineffective in updating the location of the gaze;therefore head movements have to be adapted to become theprime mover for directing gaze and executing scanning. MostSPV investigators have adopted a set up where the image issourced from a head-mounted camera such that head movementswill necessarily re-center the gaze of the phosphene field, or haveincorporated a head tracker device to update computer-generatedvirtual scenes. Eye-directed scanning is still available to the recip-ients of subretinal based implants (e.g. Wilke et al., 2009; Zrenneret al., 2006, 2007) or for device designs that seek to incorporate aneye tracker; therefore, eye-directed scanning behavior continues tobe of interest.

4.2. Visual scanning in SPV

There is substantial support from SPV studies that restraint-freevisual scanning has a positive impact on the task performance byallowing subjects self-directed visual information maximization.Cha and co-workers (1992a) suggested that their subjects directedgaze so as to ‘‘choose” the best viewing position in order to identifythe orientation of the Tumbling E, thus improving LA scores. Sim-ilarly, with irregular phosphene matrices, Cai and co-workers(2005) noted that subjects learned to select areas with denserphosphenes for scanning in order to maximize the spatial fre-quency content of the stimuli. Head movements also expand theeffective field of view and allow rudimentary depth perceptionthrough parallax (Cha et al., 1992c).

With respect to scanning strategies, Dagnelie and co-workers(2006b), in a study involving a hand-eye coordination task, notedthat different scanning approaches between subjects essentiallyafforded them the same performance outcome; but no formal com-parison was made. Other investigators found that performance washighly influenced by the scanning strategy.


In an LA task with the Landolt C optotype, Chen and co-workers(2006) showed that subjects who had adopted a circular scanningstrategy scored consistently and significantly better than others. Inaddition, they also showed a weak, positive correlation betweenscanning speed and performance. Hayes and co-workers (2003)also remarked that during a reading task, two of their subjectswere able to read finer text due to the fast scanning techniquesthey adopted; presumably, the fast temporal changes in the phos-phenes compensated for the low phosphene resolution. Humayunand co-workers (2001) also arrived at the same conclusion, observ-ing that proper scanning technique and scanning velocity can im-prove upon SPV face recognition, object discrimination andreading.

In a task requiring SPV reading of paragraphed text using com-bined eye and head-directed gaze, Sommerhalder and co-workers(Sommerhalder et al., 2004, Perez Fornos et al., 2006) also foundthat particular scanning actions contributed significantly to theability to read text more promptly. They observed that initially,eye movements of subjects were littered with ineffective saccadesand reflexive actions. Gradually the subjects improved control andstability over the vertical eye movements so as to suppress thereflexive actions that attempts to refocus the text to the fovearather than the eccentric viewing window. This was comple-mented by a more controlled and effective horizontal scanningeye movement of the eccentric viewing window, and entailedimprovement in the reading speeds.

In a separate study requiring subjects to locate a light sourceand perform a pattern matching task, Perez Fornos and co-workers (2008) also made interesting observations of the eyeand head movements of their subjects. They showed that theamount of head and eye movements was negatively correlatedto the size of the phosphene field. ‘‘Tunnel vision” with smallerphosphene fields has meant that lengthier visual search has tobe performed.

Sommerhalder and co-workers’ studies (Sommerhalder et al.,2003, 2004, Perez Fornos et al., 2005, 2008) were conducted withretinal-stabilized phosphenes. Under this experimental paradigm,phosphene locations are fixed with respect to their visual fieldcoordinates. Retinal-stabilized phosphenes mimic physiologicalphosphenes which appear to move with eye movements, verymuch like the afterimages subsequent to staring at a bright lightsource. However, the retinal-stabilized protocol has not been com-monly adopted in SPV studies.

A few studies investigated the performance difference betweenretinal-stabilized phosphene simulations and non-stabilized simu-lations. Cha et al. (1992b) compared the reading speed of head-di-rected and eye-directed (retinal-stabilized) scanning under thesame SPV configuration and noted that eye-directed scanning pro-duced poorer results than head-directed scanning. On the otherhand, in a study involving a hand-eye coordination task througha retinal-stabilized phosphene field3 (Dagnelie, Barnett, et al.,2006), it was found that though initially the performance in the taskdeclined compared to previous non-stabilized sessions, subjectsquickly learnt to keep the eyes steady and scan with head move-ments. Subsequently the performance rose to a level similar to thefree viewing condition. It appears that eye-directed scanning led tolower task performance, but this can be compensated with appropri-ate head scanning adaptation.

Wang and co-workers (2008) examined the ability of prostheticvision recipients in conducting the smooth pursuit action using eyemovements. In this study, a slow-moving target has to be trackedusing a phosphene field, the position of which was controlled by

3 Dagnelie and co-workers’ (2006a) phosphene field was only ‘‘approximately”retinal-stabilized as eye tracking was performed only at 30 Hz.

eye movements. It was found that both latency to movement onsetand accuracy of tracking was reduced in prosthetic vision (com-pared to normal vision) and further decreased if tracking had tobe performed over an eccentric retinal locus. Interestingly, theynoted that as opposed to a ‘‘smooth” pursuit, most subjects’ track-ing locus exhibited stepwise saccades. This behavior indicates thatthe customary pursuit eye movement may not be the most effi-cient for object tracking in a visual scene composed of discretephosphenes.

In summary, it should be noted that the adaptation to a trans-formed visual input requires the adaptation of motor actions aswell as the adaptation of perception – these are two different phe-nomena. Part of the adaptation is to adopt a new set of ‘‘rules” forlooking and acting, for how the visual system now has to work inconjunction with the motor system, such as in visual scanning orcoordinating head movements with intended limb movements,etc. Chen and co-workers (2006) reported that only 33% of thetwelve subjects participating in an LA task managed to recognize(consciously or subconsciously) the higher efficacy of and adaptto the circular scanning strategy. Therefore training in the coordi-nation of visual input and motor actions should play a central partin any prosthesis recipient rehabilitation program.

5. Relating simulations to real prosthetic vision

SPV opens the opportunity to study functional prosthetic visionwhere and when real prosthetic vision recipients do not exist or arenot available. Functional prosthetic vision can be assessed in con-trolled, real or virtual-reality environments in which scientific obser-vations can be conducted. The main purpose of SPV is to betterunderstand the utility of real prosthetic vision. For this reason, inves-tigations in SPV should focus on relevant issues encountered byresearchers and,where required, lead the research community infind-ing the solution. Three of the most pressing questions at present are:

1. What are the design requirements for functional prostheticvision?

2. How can prosthetic recipients make the best use of the currentgeneration of low-phosphene-count implants?

3. How should the image/video be processed to present the mostcomprehensible prosthetic vision?

5.1. Design requirements for functional prosthetic vision

Most visual prosthesis researchers would misquote Cha and co-worker’s (1992a, 1992b, 1992c) estimate of 625 (25 � 25) phosph-enes as the requirement for close-to-normal performance in LA,text reading and mobility. However, they fail to indicate that theseare 625 phosphenes over a compact 1.7� � 1.7� visual field. Inwhich case, Cha and co-workers’ finding should be more appropri-ately interpreted as pertaining to the optimum phosphene densityor PS of a phosphene array, rather than the phosphene count; theperformance of Cha and co-workers’ phosphene fields for LA andreading at the same phosphene density but different phosphenecount was statistically equal.

As reviewed earlier, performance in perception of detail ismainly a function of PS. The trend fitted to the data in SPV studiesin Fig. 4 serves as a first approximation of the expected perfor-mance in prosthetic vision recipients. From the figure, equivalenceto normal LA can be attained with a PS of 2.50.

From the viewpoint of affording implant recipients the ability toread, SPV studies (Cha et al., 1992b; Dagnelie, Walter, et al., 2006;Fu et al., 2006) indicate that:

� The phosphene field has to contain a window of 2–4 characterswide and two lines of text high to facilitate paragraph reading.


� 2.5–4 phosphenes per character per axis are required for anyappreciable reading speed and text comprehension.

Using this information, the minimum requirement for readingwith the text occupying the entire phosphene field is likely to bearound 12 � 8 (96) phosphenes. In general, providing the readingwindow remains unchanged, the reading speed will increase asthe number of phosphenes increases. Low vision patients can af-ford to read with LA reserve of 3:1, giving 36 � 24 (864) phosph-enes. Optimum reading speed occurs at 6:1 to 18:1 LA reserve innormally-sighted individuals (Whittaker & Lovie-Kitchin, 1993),which corresponds to a phosphene field of up to 216 � 144(31,104) phosphenes – at these phosphene resolutions, most dailytasks should also be achievable.

For navigation, the visual field size of the phosphene field is animportant factor. A visual field area of at least 30� (Cha et al.,1992c) and up to an eccentricity of 58� (Lovie-Kitchin et al.,1990) would be required for self-orientation and safe navigation.However, enlarging the visual field coverage comes at the expenseof losing LA if the number of electrodes (phosphenes) is fixed. LAis important for path finding and obstacle navigation; therefore itis probably most appropriate to mimic the human retina byretaining a central high acuity phosphene field with peripheralplacement of additional phosphenes at lesser acuity to assist nav-igation. As a first approximation, ten to twenty peripheral phos-phenes to each side of the horizontal meridian and the inferiordivision of the visual field may suffice for most orientation andmobility activities.

Needless to say, these specifications remain entirely theoreticalat present. In the advent of more human trials, the clinical data willlead to improvements in SPV configurations and assessment tech-niques, and through these, the models developed will enable betterdesign of future visual prosthesis devices.

4 Hallum and co-workers (2004b) used an averaging kernel shaped in a regularhexagon, where r indicates the distance from the centre of the hexagon to its vertex.

5.2. Making the most from low phosphene count

Most of the SPV studies were conducted with configurationsthat were too optimistic compared to that of near-term visualprosthesis devices. The chronic implants in human trials at presentconsist either of 4 � 4 epiretinal arrays (Humayun et al., 2003),6 � 10 epiretinal arrays (Ahuja et al., 2009; McMahon et al.,2009), or a 4 electrode optic nerve cuff (Veraart et al., 1998, ableto elicit 100 + individual phosphenes). There are several other hu-man trials in planning, but with similar counts of available elicit-able phosphenes expected.

The resolution of these devices is in the order of 1–2� visual an-gle per phosphene. This is equivalent to a LA of 2.0 logMAR orworse, and by definition, still classified as legally blind (>1.0 log-MAR LA). Hayes and co-workers (2003) demonstrated that basicobject and symbol discrimination is possible with 4 � 4 and6 � 10 phosphenes. Using a similar configuration, Hallum and co-workers (Hallum et al. 2003, 2005) demonstrated the ability totrack object motion. Visually guided tasks affordable by these hu-man trial devices corresponded to these observations from SPVinvestigations, namely simple light localization, motion detectionand simple pattern discrimination (Brelen, Duret, Gerard, Delbeke,& Veraart, 2005; Duret et al., 2006; Veraart, Wanet-Defalgue, Ger-ard, Vanlierde, & Delbeke, 2003; Yanai et al., 2007).

It appears that the more difficult tasks of reading, face recogni-tion, and navigation may be impractical for recipients of near-termvisual prosthesis. Yet in SPV studies, using only 4 � 4 and 6 � 10phosphenes, subjects were able to pour candy from one cup intoanother without spillage (Hayes et al., 2003), or navigate a simpleobstacle course (Dagnelie et al., 2007). Therefore, theoretically, alow-count phosphene field is not entirely useless for more compli-

cated daily tasks. More investigations are required to rectify theperformance differences between SPV studies and real prostheticvision – whether this is due to a learning or training factor, irreg-ularity of the phosphene map, inappropriate simulation orotherwise.

Perceptual learning and the development of scanning skills ap-pear to play an important role in helping subjects in SPV to im-prove upon their response correctness and response time. Thebenefits of visual scanning have also been implied in several hu-man trials. Particularly, a head-directed visual scanning routinebased on scanning in the horizontal direction followed by the ver-tical for identifying symbols using an optic nerve implant wasdeveloped (Brelen et al., 2005; Duret et al., 2006; Veraart et al.,2003). With scanning, Dobelle’s (2000) cortically implanted volun-teer also could routinely recognize letters and symbols, and furtherbelieved that with additional experience in scanning, performancein other tasks will also improve.

Rather than an afterthought, SPV investigators can designexperiments for establishing a set of practical scanning strategiesunder different usage scenarios: for conducting a visual search,tracking an object, text reading, and navigation, etc. For example,scanning strategy would be particularly useful to help the pros-thetic vision recipient better follow the lines in the visual scene,such as junctions between the wall and the floor or line of lightsin the ceiling, for orientation and mobility tasks (Dobelle, 2000).

5.3. Image processing for visual comprehension

To represent a scene captured by a camera, which is normally ofat least 640 � 480 resolution, with a field of 100 or so phosphenesnaturally requires some degree of image processing. In the case ofcochlear implants, it is estimated that a third of the improvementin the monosyllabic word recognitions from the 1980s to 1999 canbe attributed to speech processing improvements (Rubinstein &Miller, 1999). Much of the speech processing work derives frompsychophysical studies with simulated models of human auditoryperception. The optimal processing strategy is yet unclear for phos-phene vision, and deserves a separate review. Here, the present re-view focuses on the very limited number of SPV studies thatinvestigated image processing strategies.

Most of the present SPVs are based on a regional average deci-mation of the full-resolution image into values that modulate theluminance or the size of the phosphenes. The circular regionalaverage was investigated by Chen and co-workers (2004, 2005b).As described earlier, they showed that averaging over a circulararea with the radius at approximately 50% PS is optimal for LA per-formance on the Landolt C optotype.

Based on the results of their smooth pursuit study, Hallum andco-workers (2004b) examined the optimum sizes for a regionalaverage kernel4 and a Gaussian filtering kernel using mutual infor-mation. The estimated optimum regional averaging kernel covers alarge pixel area (r = 180% PS), and the optimum Gaussian kernel alsohas a relatively large spread (r = 85% PS). Overall, the Gaussian ker-nel is shown to convey more information than the regional averagekernel. The larger ‘‘input apertures” estimated appear to be optimalspecifically for movement detection which is the essence of thesmooth pursuit task. In addition to mutual information, Hallumand co-workers (2006a, 2006b) have also used techniques basedon Fourier analysis to investigate the stochastic fields of phosphenesand the performance impact of inaccurate phosphene visual field po-sition mapping.


A different approach to image processing was proposed byEckmiller and co-workers (Baruth, Eckmiller, & Neumann, 2003;Becker, Eckmiller, & Hunermann, 1999; Eckmiller, Baruth, &Neumann, 2004). Their concept is to use an artificial neuralnetwork to convert a full-resolution video into spike patterns forsimulating retinal ganglion cells – the ‘‘tunable retinal encoder”.They have described psychophysical experiments where humansubjects helped the artificial neural network to learn the optimalconversion strategy. However, these techniques have not yet beenrigorously applied in SPV studies of performance.

While an analytic approach to assess the functional capacity ofprosthetic vision can provide a general estimate of the required im-age processing routines, psychophysical studies with human sub-jects provide the best platform to help select from among a setof image processing strategies the one best suited to facilitate func-tional vision, for a particular daily task. This information can thenbe fed back into a prosthetic vision performance model for furtheranalytical research. There is certainly no shortage of image pro-cessing techniques available for consideration – spatial filtering,edge detection, contrast enhancement, saliency mapping, etc. Con-sidering the low resolution of near-term devices, perhaps investi-gators should focus on processing strategies that stimulatetemporal processing pathways in the human visual system forcompensating the lack of spatial resolution.

5.4. Towards a synergistic approach to SPV

In the research field of cochlear implants, several related stan-dard measures of hearing performance have been established.The measure of monosyllabic words seems to have established it-self as a pseudo-reference in clinical assessments; extended mea-sures involve recognition of vowels, consonants, words andsentences. Having referential measures allows comparison be-tween speech processing algorithms, implant technology, etc. Incontrast, presently, each SPV investigator would set about investi-gating the functional capacity of prosthetic vision in their individ-ual ways.

The LA metric has the potential to act as a reference measureto improve comparability between SPV study results across dif-ferent investigator groups all over the world. Not only resultsfrom similar psychophysical tasks can be compared against oneanother, having a reference measure means that interpretationcan be further extrapolated to performance in other tasks. Aspresented in this review, the LA metric is a good indicator ofthe ability to appreciate the detailed information in a visualscene and can be a good predictor of many visually guided tasksin daily activities. Likewise, for prosthetic vision, the more highspatial frequency information a recipient of a visual prosthesissystem can resolve, the better the visual quality and functionalvision the recipient can enjoy.

An LA reference measurement can be made very efficientlyusing standard clinical procedures by assessing subjects with vari-ants of letter charts. To use LA as a means for modeling functionalperformance or for statistical comparisons, a more time consumingprotocol is recommended to reduce bias and variance in themeasurements5.

5 In the authors’ experience using a computerized, automated test setup, asufficient measurement can be obtained with 10–20 test items at eight optotype sizesof stepping 0.1 logMAR with the projected LA threshold approximately in the centerof this size range. Experienced subjects are expected to take between 1 and 3 s torespond per test item; within this time frame, experienced subjects can typicallyperform 1–2 circular scans over the item. The entire exercise would not exceed10 min.

� Use optotypes of a single character, such as the Tumbling E andthe Landolt C, rather than letters of the alphabet where each hasa different recognition difficulty.6

� Assess using the method of constant stimuli.� Estimate LA threshold by fitting a psychometric function to the

percentage performance versus optotype size (in logMAR) whichalso provides statistical information regarding the estimated LAthreshold.

� Assessment optotype sizes should cover the range from near100% recognition to recognition by chance to give an accuratepsychometric fitting.

� Observe the learning effect: Allow four to six sessions of prac-tice, each on a separate day.

� Facilitate natural head and eye scanning actions to maximize theperformance outcome. Monitor and record the scanning behav-ior for analysis.

� Detailed report of the SPV configuration to facilitate comparisonby other investigators.

Of course, no single measure can characterize the functionalneeds for every task. LA, by itself, is assessed on localized, high con-trast test items, and is not always the strongest predictor in everytask. In addition to LA, clinical measures of contrast sensitivity(Pelli-Robson or Regan charts) also correlate significantly withmany daily activities, and the requirement of the visual field sizeis important with respect to the size of the reading window andspatial orientation in navigation tasks. The SPV research commu-nity should work together to identify a small set of reference mea-surements that characterizes the performance in key functionaltasks and can be easily standardized across laboratories forinteroperability.

When used effectively, psychophysics under SPV can be a veryinformative tool for studying the functional capacity of realprosthetic vision. Through SPV, the investigators can experiencefirst-hand many performance and behavioral aspects of prostheticvision, and studies conducted on a larger population can inform theperformance and behavioral preferences in general and in individ-ual cases. A human performance model of prosthetic vision can beconstructed to supplement the analytical perspective of the re-search. A visual prosthesis device and image processing strategycan be designed to target the areas of functional living where itmatters most. Behavioral findings can be compiled so as to engi-neer a successful and effective prosthetic vision rehabilitation pro-gram (for example, Chen, Suaning, Morley, & Lovell, 2009b;Dagnelie, 2008). The success or failure of a visual prosthesis systemmay ultimately depend on how useful implant recipients find theirnewly bestowed prosthetic vision.

References

Ahuja, A. K., Dorn, J. D., Caspi, A., Filley, E., Dagnelie, G., Greenberg, R. J., et al. (2009).The Argus II retinal prosthesis enables blind subjects to identify the direction ofmotion. Investigative Ophthalmology & Visual Science, 50 (E-Abstract 4590).

Akutsu, H., Bedell, H. E., & Patel, S. S. (2000). Recognition thresholds for letters withsimulated dioptric blur. Optometry & Vision Science, 77, 524–530.

Baruth, O., Eckmiller, R., & Neumann, D. (2003). Retina encoder tuning and dataencryption for learning retina implants. In Proceedings of the international jointconference on neural networks 2003. (Vol. 2, pp. 1249–1252).

Becker, M., Eckmiller, R., & Hunermann, R. (1999). Psychophysical test of a tunableretina encoder for retina implants. In IJCNN 1999: International joint conferenceon neural networks (Vol. 1, pp. 192–195).

Boisjoly, H., Gresset, J., Fontaine, N., Charest, M., Brunette, I., Lefrancois, M., et al.(1999). The VF-14 index of functional visual impairment in candidates for acorneal graft. American Journal of Ophthalmology, 128, 38–44.

6 More complex optotypes can be used for LA assessment once their performancerelationships are established for prosthetic vision such that measurements fromdifferent studies are comparable.


Bowers, A. R., & Reid, V. M. (1997). Eye movements and reading with simulatedvisual impairment. Opthalmic and Physiological Optics, 17, 392–402.

Boyle, J., Maeder, A., & Boles, W. (2001). Static image simulation of electronic visualprostheses. In The Seventh Australian and New Zealand intelligent informationsystems conference, 2001 (pp. 85–88).

Boyle, J., Maeder, A., & Boles, W. (2002a). Visual perception of low quality images. InICONIP 2002: Proceedings of the 9th international conference on neural informationprocessing (Vol. 1, pp. 153–157).

Boyle, J., Maeder, A., & Boles, W. (2003). Scene specific imaging for bionic visionimplants. In ISPA 2003: Proceedings of the 3rd international symposium on imageand signal processing and analysis (Vol. 1, pp. 423–427).

Boyle, J. R., Maeder, A. J., & Boles, W. W. (2002b). Image enhancement for electronicvisual prostheses. Journal of Australasian College of Physical Scientists andEngineers in Medicine, 25, 81–86.

Brelen, M., Duret, F., Gerard, B., Delbeke, J., & Veraart, C. (2005). Creating ameaningful visual perception in blind volunteers by optic nerve stimulation.Journal of Neural Engineering, 2, S22–S28.

Brindley, G. S., & Lewin, W. S. (1968). The sensations produced by electricalstimulation of the visual cortex. Journal of Physiology, 196, 479–493.

Brindley, G. S., & Rushton, D. N. (1974). Implanted stimulators of the visual cortex asvisual prosthetic devices. Transactions of the American Academy ofOphthalmology and Otolaryngology, 78, OP741–OP745.

Brown, B., Brabyn, L., Welch, L., Haegerstrom-Portnoy, G., & Colenbrander, A.(1986). Contribution of vision variables to mobility in age-relatedmaculopathy patients. American Journal of Optometry & PhysiologicalOptics, 63, 733–739.

Bullimore, M. A., & Bailey, I. L. (1995). Reading and eye movements in age-relatedmaculopathy. Optometry & Vision Science, 72, 125–138.

Bullimore, M. A., Bailey, I. L., & Wacker, R. T. (1991). Face recognition inage-related maculopathy. Investigative Ophthalmology & Visual Science, 32,2020–2029.

Button, J., & Putnam, T. (1962). Visual responses to cortical stimulation in the blind.Journal of the Iowa Medical Society, 52, 17–21.

Cai, S., Fu, L., Zhang, H., Hu, G., & Liang, Z. (2005). Prosthetic visual acuity in irregularphosphene arrays under two down-sampling schemes: A simulation study. InIEEE-EMBS 2005: 27th annual international conference of the IEEE Engineering inMedicine and Biology Society (pp. 5223–5226).

Carta, A., Braccio, L., Belpoliti, M., Soliani, L., Sartore, F., Gandolfi, S. A., et al. (1998).Self-assessment of the quality of vision: association of questionnaire score withobjective clinical tests. Current Eye Research, 17, 506–512.

Cha, K., Horch, K., & Normann, R. (1992a). Simulation of a phosphene-based visualfield: Visual acuity in a pixelized vision system. Annals of Biomedical Engineering,20, 439–449.

Cha, K., Horch, K., Normann, R., & Boman, D. (1992b). Reading speed with a pixelizedvision system. Journal of the Optical Society of America A Optics and Image Science,9, 673–677.

Cha, K., Horch, K. W., & Normann, R. A. (1992c). Mobility performance with apixelized vision system. Vision Research, 32, 1367–1372.

Chai, X., Yu, W., Wang, J., Zhao, Y., Cai, C., & Ren, Q. (2007). Recognition of pixelizedChinese characters using simulated prosthetic vision. Artificial Organs, 31,175–182.

Chai, X., Zhang, L., Li, W., Shao, F., Yang, K., & Ren, Q. (2008). Study of tactileperception based on phosphene positioning using simulated prosthetic vision.Artificial Organs, 32, 110–115.

Chen, S. C., Hallum, L. E., Lovell, N. H., & Suaning, G. J. (2005a). Learning prostheticvision: A virtual-reality study. IEEE Transactions on Neural Systems andRehabilitation Engineering, 13, 249–255.

Chen, S. C., Hallum, L. E., Lovell, N. H., & Suaning, G. J. (2005b). Visual acuitymeasurement of prosthetic vision: A virtual-reality simulation study. Journal ofNeural Engineering, 2, S135–S145.

Chen, S. C., Chen, S. C., Hallum, L. E., Suaning, G. J. & Lovell, N. H. (2006).Psychophysics of prosthetic vision: I. Visual scanning and visual acuity. In IEEE-EMBS 2006: 28th annual international conference of the IEEE Engineering inMedicine and Biology Society (pp. 4400–4403).

Chen, S. C., Hallum, L. E., Suaning, G. J., & Lovell, N. H. (2007). A quantitative analysisof head movement behaviour during visual acuity assessment under prostheticvision simulation. Journal of Neural Engineering, 4, S108–S123.

Chen, S. C., Lovell, N. H., & Suaning, G. J. (2004) Effect on prosthetic vision visualacuity by filtering schemes, filter cut-off frequency and phosphene matrix: avirtual reality simulation. In IEEE-EMBS 2004: 26th annual internationalconference of the IEEE Engineering in Medicine and Biology Society (Vol. 2, pp.4201–4204).

Chen, S. C., Suaning, G. J., Morley, J. W., & Lovell, N. H. (2009a). Simulatingprosthetic vision: I. Visual models of phosphenes. Vision Research, 49,1493–1506.

Chen, S. C., Suaning, G. J., Morley, J. W., & Lovell, N. H. (2009b). Rehabilitationregimes based upon psychophysical studies of prosthetic vision. Journal ofNeural Engineering, 6, 035009.

Costen, N. P., Parker, D. M., & Craw, I. (1994). Spatial content and spatialquantisation effects in face recognition. Perception, 23, 129–146.

Dagnelie, G. (2008). Psychophysical evaluation for visual prosthesis. Annual Reviewof Biomedical Engineering, 10, 339–368.

Dagnelie, G., Barnett, D., Humayun, M. S., & Thompson, R. W. Jr., (2006a). Paragraphtext reading using a pixelized prosthetic vision simulator: Parameterdependence and task learning in free-viewing conditions. InvestigativeOphthalmology & Visual Science, 47, 1241–1250.

Dagnelie, G., Keane, P., Narla, V., Yang, L., Weiland, J., & Humayun, M. (2007). Realand virtual mobility performance in simulated prosthetic vision. Journal ofNeural Engineering, 4, S92–S101.

Dagnelie, G., Walter, M., & Liancheng, Y. (2006b). Playing checkers: Detection andeye-hand coordination in simulated prosthetic vision. Journal of Modern Optics,53, 1325–1342.

Delbeke, J., Wanet-Defalque, M. C., Gerard, B., Troosters, M., Michaux, G., & Veraart,C. (2002). The microsystems based visual prosthesis for optic nerve stimulation.Artificial Organs, 26, 232–234.

Dixon, P., & Dilollo, V. (1994). Beyond visible persistence: An alternative account oftemporal integration and segregation in visual processing. Cognitive Psychology,26, 33–63.

Dobelle, W. H. (2000). Artificial vision for the blind by connecting a televisioncamera to the visual cortex. American Society for Artificial Internal Organs, 46,3–9.

Dobelle, W. H., Mladejovsky, M. G., & Evans, J. R. (1976). ‘Braille’ reading by a blindvolunteer by visual cortex stimulation. Nature, 259, 111–112.

Dobelle, W. H., Mladejovsky, M. G., & Girvin, J. P. (1974). Artificial vision for theblind: electrical stimulation of visual cortex offers hope for a functionalprosthesis. Science, 183, 440–444.

Dowling, J., Boles, W. & Maeder, A. (2005). Mobility assessment using simulatedartificial human vision. In 2005 IEEE computer society conference on computervision and pattern recognition (Vol. 3, pp. 32–32).

Dowling, J. A., Maeder, A., & Boles, W. (2004). Mobility enhancement, assessment fora visual prosthesis. Proceedings of SPIE: Medical Imaging Physiology, Function, andStructure from Medical Images, 5369, 780–791.

Duret, F., Brelen, M. E., Lambert, V., Gerard, B., Delbeke, J., & Veraart, C. (2006).Object localization, discrimination, and grasping with the optic nerve visualprosthesis. Restorative Neurology and Neuroscience, 24, 31–40.

Eckmiller, R., Baruth, O., & Neumann, D. (2004). Neural information processing effortsto restore vision in the blind. Neural information processing (Vol. 3316, pp. 10–18).Berlin/Heidelberg: Springer.

Fu, L., Cai, S., Zhang, H., Hu, G., & Zhang, X. (2006). Psychophysics of reading with alimited number of pixels: Towards the rehabilitation of reading ability withvisual prosthesis. Vision Research, 46, 1292–1301.

Geruschat, D. R., Turano, K. A., & Stahl, J. W. (1998). Traditional measures of mobilityperformance and retinitis pigmentosa. Optometry & Vision Science, 75, 525–537.

Gold, J., Bennett, P. J., & Sekuler, A. B. (1999). Identification of band-pass filteredletters and faces by human and ideal observers. Vision Research, 39, 3537–3560.

Hallum, L. E., Chen, S. C., Cloherty, S. L. & Lovell, N. H. (2006). Psychophysics ofprosthetic vision: II. Stochastic sampling, the phosphene image, and noise. InIEEE-EMBS 2006: 28th annual international conference of the IEEE Engineering inMedicine and Biology Society (pp. 1634–1637).

Hallum, L. E., Cloherty, S. L., Taubman, D. S., Suaning, G. J. & Lovell, N. H. (2006).Psychophysics of prosthetic vision: III. Stochastic rendering, the phospheneimage, and perception. In IEEE-EMBS 2006: 28th annual international conferenceof the IEEE Engineering in Medicine and Biology Society (pp. 1169–1172).

Hallum, L. E., Dagnelie, G., Suaning, G. J., & Lovell, N. H. (2007). Simulating auditoryand visual sensorineural prostheses: A comparative review. Journal of NeuralEngineering, 4, S58–S71.

Hallum, L. E., Suaning, G. J., & Lovell, N. H. (2004a). Contribution to the theory ofprosthetic vision. ASAIO Journal, 50, 392–396.

Hallum, L. E., Suaning, G. J., Taubman, D. S. & Lovell, N. H. (2004). Towardsphotosensor movement-adaptive image analysis in an electronic retinalprosthesis. In IEEE-EMBS 2004: 26th annual international conference of the IEEEEngineering in Medicine and Biology Society (Vol. 2, pp. 4165–4168).

Hallum, L. E., Suaning, G. J., Taubman, D. S., & Lovell, N. H. (2005). Simulatedprosthetic visual fixation, saccade, and smooth pursuit. Vision Research, 45,775–788.

Hallum, L. E., Taubman, D., Suaning, G. J., Morley, J. & Lovell, N. H. (2003). A filteringapproach to artificial vision: a phosphene visual tracking task. In IFMBEproceedings, World congress on medical physics and biomedical engineering,August 24–29, Sydney.

Haymes, S., Guest, D., Heyes, A., & Johnston, A. (1996). Mobility of people withretinitis pigmentosa as a function of vision and psychological variables.Ophthalmic and Physiological Optics, 73, 621–637.

Haymes, S. A., Johnston, A. W., & Heyes, A. D. (2002). Relationship between visionimpairment and ability to perform activities of daily living. Ophthalmic andPhysiological Optics, 22, 79–91.

Hayes, J. S., Yin, V. T., Piyathaisere, D., Weiland, J. D., Humayun, M. S., & Dagnelie, G.(2003). Visually guided performance of simple tasks using simulated prostheticvision. Artificial Organs, 27, 1016–1028.

Hazel, C. A., Petre, K. L., Armstrong, R. A., Benson, M. T., & Frost, N. A. (2000). Visualfunction and subjective quality of life compared in subjects with acquiredmacular disease. Investigative Ophthalmology and Visual Science, 41,1309–1315.

Humayun, M. S. (2001). Intraocular retinal prosthesis. Transactions of the AmericanOphthalmological Society, 99, 271–300.

Humayun, M. S., Weiland, J. D., Fujii, G. Y., Greenberg, R., Williamson, R., Little, J.,et al. (2003). Visual perception in a blind subject with a chronic microelectronicretinal prosthesis. Vision Research, 43, 2573–2581.

Humayun, M., Yanai, D., Greenberg, R. J., Little, J., Mech, B. V., Mahadevappa, M.,et al. (2004). Clinical results with the model 1IRP implant. In 2004 IEEEinternational joint conference on neural networks.

Hungenahally, S. (1995). Mathematical basis for the design of an artificial retina:visual prosthesis for the retinally blind. In ‘Intelligent Systems for the 21st


Century’., 1995 IEEE international conference on systems, man and cybernetics (Vol.3, pp. 2396–2401).

Jan, J. E. (1991). Head movements of visually impaired children. DevelopmentMedicine and Child Neurology, 33, 645–647.

Karni, A., & Sagi, D. (1993). The time course of learning a visual skill. Nature, 365,250–252.

Klein, R., Moss, S. E., Klein, B. E. K., Gutierrez, P., & Mangione, C. M. (2001). The NEI-VFQ-25 in people with long-term type 1 diabetes mellitus: The Wisconsinepidemiologic study of diabetic retinopathy. Archives of Ophthalmology, 119,733–740.

Legge, G. E., Pelli, D. G., Rubin, G. S., & Schleske, M. M. (1985). Psychophysics ofreading – I. Normal vision. Vision Research, 25, 239–252.

Lovie-Kitchin, J., Mainstone, J., Robinson, J., & Brown, B. (1990). What areas of thevisual field are important for mobility in low vision patients? Clinical VisionScience, 5, 249–263.

Mangione, C. M., Gutierrez, P. R., Lowe, G., Orav, E. J., & Seddon, J. M. (1999).Influence of age-related maculopathy on visual functioning and health-relatedquality of life. American Journal of Ophthalmology, 128, 45–53.

Mangione, C. M., Lee, P. P., Gutierrez, P. R., Spritzer, K., Berry, S., & Hays, R. D. (2001).Development of the 25-item national eye institute visual functionquestionnaire. Archives of Ophthalmology, 119, 1050–1058.

Mcclure, M. E., Hart, P. M., Jackson, A. J., Stevenson, M. R., & Chakravarthy, U. (2000).Macular degeneration: Do conventional measurements of impaired visualfunction equate with visual disability? British Journal of Ophthalmology, 84,244–250.

Mcmahon, T. T., Hansen, M., & Viana, M. (1991). Fixation characteristics in maculardisease. Relationship between saccadic frequency, sequencing, and reading rate.Investigative Ophthalmology & Visual Science, 32, 567–574.

Mcmahon, M. J., Dorn, J. D., Ahuja, A. K., Caspi, A., Filley, E., Dagnelie, G. & Greenberg,R. J. (2009). The Argus II retinal prosthesis enables blind subjects to localizeobjects. Investigative Ophthalmology & Visual Science, 50 (E-Abstract 4589).

Nasanen, R. (1999). Spatial frequency bandwidth used in the recognition of facialimages. Vision Research, 39, 3824–3833.

Pérez Fornos, A., Sommerhalder, J., Pittard, A., Safran, A. B., & Pelizzone, M. (2008).Simulation of artificial vision: IV. Visual information required to achieve simplepointing and manipulation tasks. Vision Research, 48, 1705–1718.

Pérez Fornos, A., Sommerhalder, J., Rappaz, B., Pelizzone, M., & Safran, A. B. (2006).Processes involved in oculomotor adaptation to eccentric reading. InvestigativeOphthalmology and Visual Science, 47, 1439–1447.

Pérez Fornos, A., Sommerhalder, J., Rappaz, B., Safran, A. B., & Pelizzone, M. (2005).Simulation of artificial vision, III: Do the spatial or temporal characteristics ofstimulus pixelization really matter? Investigative Ophthalmology & VisualScience, 46, 3906–3912.

Richard, G., Feucht, M., Bornfeld, N., Laube, T., Rossler, G., Velikay-Parel, M. & Hornig,R. (2005). Multicenter study on acute electrical stimulation of the human retinawith an epiretinal implant: Clinical results in 20 patients. InvestigativeOphthalmology & Visual Science, 46 (E-Abstract 1143).

Richard, G., Hornig, R., Keseru, M. & Feucht, M. (2007). Chronic epiretinal chipimplant in blind patients with retinitis pigmentosa: Long-term clinical results.Investigative Ophthalmology & Visual Science, 48 (E-Abstract 666).

Rubinstein, J. T., & Miller, C. A. (1999). How do cochlear prostheses work? CurrentOpinion in Neurobiology, 9, 399–404.

Rushton, D. N., & Brindley, G. S. (1978). Properties of cortical electrical phosphenes.In S. J. Cool & E. J. I. Smith (Eds.), Frontiers in visual science. New York: Springer-Verlag.

Sagi, D., & Tanne, D. (1994). Perceptual learning: Learning to see. Current Opinion inNeurobiology, 4, 195–199.

Solomon, J. A., & Pelli, D. G. (1994). The visual filter mediating letter identification.Nature, 369, 395–397.

Sommerhalder, J., Oueghlani, E., Bagnoud, M., Leonards, U., Safran, A. B., & Pelizzone,M. (2003). Simulation of artificial vision: I. Eccentric reading of isolated words,and perceptual learning. Vision Research, 43, 269–283.

Sommerhalder, J., Rappaz, B., De Haller, R., Fornos, A., Safran, A. B., & Pelizzone, M.(2004). Simulation of artificial vision: II. Eccentric reading of full-page text andthe learning of this task. Vision Research, 44, 1693–1706.

Szlyk, J. P., Fishman, G. A., Alexander, K. R., Revelins, B. I., Derlacki, D. J., & Anderson,R. J. (1997). Relationship between difficulty in performing daily activities andclinical measures of visual function in patients with retinitis pigmentosa.Archives of Ophthalmology, 115, 53–59.

Szlyk, J. P., Seiple, W., Fishman, G. A., Alexander, K. R., Grover, S., & Mahler, C. L. (2001).Perceived and actual performance of daily tasks: Relationship to visual functiontests in individuals with retinitis pigmentosa. Ophthalmology, 108, 65–75.

Thompson, R. W., Barnett, G. D., Humayan, M., & Dagnelie, G. (2003). Facialrecognition using simulated prosthetic pixelized vision. InvestigativeOphthalmology & Visual Science, 44, 5035–5042.

Velikay-Parel, M., Ivastinovic, D., Koch, M., Hornig, R., Dagnelie, G., Richard, G., et al.(2007). Repeated mobility testing for later artificial visual function evaluation.Journal of Neural Engineering, 4, S102–S107.

Veraart, C., Duret, F., Brelen, M. & Delbeke, J. (2004). Vision rehabilitation with theoptic nerve visual prosthesis. In IEEE-EMBS 2004: conference proceedings. 26thannual international conference of the IEEE Engineering in Medicine and BiologySociety (Vol. 6, pp. 4163–4164).

Veraart, C., Raftopoulos, C., Mortimer, J. T., Delbeke, J., Pins, D., Michaux, G., et al.(1998). Visual sensations produced by optic nerve stimulation using animplanted self-sizing spiral cuff electrode. Brain Research, 813, 181–186.

Veraart, C., Wanet-Defalgue, M., Gerard, B., Vanlierde, A., & Delbeke, J. (2003).Pattern recognition with the optic nerve visual prosthesis. Artificial Organs, 27,996–1002.

Vurro, M., Baselli, G., Orabona, F. & Sandini, G. (2006). Simulation and assessment ofbioinspired visual processing system for epi-retinal prostheses. In IEEE-EMBS2006: 28th annual international conference of the IEEE Engineering in Medicine andBiology Society (pp. 3278–3281).

Wang, L., Yang, L., & Dagnelie, G. (2008). Initiation and stability of pursuit eyemovements in simulated retinal prosthesis at different implant locations.Investigative Ophthalmology & Visual Science, 49, 3933–3939.

Weiland, J. D., Yanai, D., Mahadevappa, M., Williamson, R., Mech, B. V., Fujii, G. Y.,et al. (2003). Electrical stimulation of retina in blind humans. In IEEE-EMBS2003: 25th annual international conference of the IEEE Engineering in Medicine andBiology Society (vol. 3, pp. 2021–2022).

Weiland, J. D., Yanai, D., Mahadevappa, M., Williamson, R., Mech, B. V., Fujii, G. Y.,et al. (2004). Visual task performance in blind humans with retinal prostheticimplants. In IEEE-EMBS 2004: 26th annual international conference of the IEEEEngineering in Medicine and Biology Society (Vol. 6, 4172–4173).

Whittaker, S. G., & Lovie-Kitchin, J. A. N. (1993). Visual requirements for reading.Optometry & Vision Science, 70, 54–65.

Wilke, R., Porubska, K., Benav, H., Kusnyerik, A., Bruckmann, A., Koch, J., et al. (2009).Visual acuity determined by landolt c test in a blind patient provided with asubretinal electronic implant. Investigative Ophthalmology & Visual Science, 50(E-Abstract 4595).

Yanai, D., Weiland, J. D., Mahadevappa, M., Greenberg, R. J., Fine, I., &Humayun, M. S. (2007). Visual performance using a retinal prosthesis inthree subjects with retinitis pigmentosa. American Journal ofOphthalmology, 143, 820–827.

Zrenner, E., Besch, D., Bartz-Schmidt, K. U., Gekeler, F., Gabel, V. P., Kuttenkeuler, C.,et al. (2006) Subretinal chronic multi-electrode arrays implanted in blindpatients. Investigative Ophthalmology & Visual Science, 47 (E-Abstract 1538).

Zrenner, E., Wilke, R., Bartz-Schmidt, K., Benav, H., Besch, D., Gekeler, F., et al. (2009).Blind retinitis pigmentosa patients can read letters and recognize the directionof fine stripe patterns with subretinal electronic implants. InvestigativeOphthalmology & Visual Science, 50 (E-Abstract 4581).

Zrenner, E., Wilke, R., Zabel, T., Sachs, H., Bartz-Schmidt, K., Gekeler, F., et al. (2007).Psychometric analysis of visual sensations mediated by subretinalmicroelectrode arrays implanted into blind retinitis pigmentosa patients.Investigative Ophthalmology & Visual Science, 48 (E-Abstract 659).

simulating prosthetic vision: ii. measuring functional ... · pdf filesimulating prosthetic...

Documents