Visual Field Recovery, Vision Restoration Therapy or simply Restitution Training (RT) are therapeutic concepts based on animal studies indicating that damaged neurons in the visual cortex can recover through repeated light stimulation (18, 19). Under the term RT exist different therapeutic concepts and methods, which range from motion discrimination of visual stimuli [e.g., (20)] to visual stimuli with different temporal and spatial frequency [e.g., (21)] and repeated visual stimulation along the area between the intact and defect visual field [e.g., (22)]. The present study focuses on the best-known therapeutic methodology of RT, namely repeated visual stimulation of the transition area between intact and defective visual fields, which is based on the “residual vision activation theory.” According to this influential concept, such reactivatable cortical areas are found, among others, along the “visual field border area” (23).

Proponents of RT argue that by stimulating the border area between the intact and damaged visual field with bright light impulses through the eyes - neurons in the corresponding (retinotopic) area of the visual cortex are reactivated - increasing the size of the intact visual field [e.g., (24, 25)]. Thus, changes after therapy with RT are perimetrically assessable [e.g., (22, 26)].¹

Some studies report considerable training effects indicating training-induced neuroplasticity in the visual cortex [e.g., (22, 26–32)]. Bergsma et al. (22) observed in their explorative study perimetrically assessable reductions of a visual field defect to varying degrees in all their 12 patients with hemianopia after RT, which was performed for 1 h a day, 5 days a week and a period of 13 weeks at home. Marshall et al. (26) show an average improvement rate of 12.5% of perimetric stimulus detection in a longitudinal cohort analysis of 7 patients with VFD after conducting RT twice daily for 20 to 30 min, 6 days a week for 3 months at home. Mueller et al. (29) report in their clinical observational study with 302 patients that RT restored up to 17.2% of their patients’ formerly blind visual fields. Results are based on clinical standard perimetry after patients conducted RT daily for 1 h, 6 days a week for 6 months at home. Poggel et al. (30) included in their pre-post intervention study without a control group 9 patients with VFD who performed RT sessions of around 15–20 min for 3 months. The authors found slight but significant improvements in light detection performance. Matteo et al. (33) conclude in their review that “[...] border rehabilitation seems to improve the detection of visual stimuli [...]” (p. 1). Bergsma et al. (22) state that “Visual restorative function training does not only lead to visual field enlargement [...] but it may also lead to subjective improvement of daily visual functioning [...]” (p. 400).

Other studies, by contrast, did not find significant effects of RT [e.g., (34–37)]. Mödden et al. (36) found in their Randomized Controlled Trial of 45 stroke patients no perimetrically assessable visual field expansion after 15 patients performed RT in 15 single sessions for 30 min for over 3 weeks. Reinhard et al. (37) performed a pre-post intervention design with 17 patients with VFD who underwent RT 1 h a day, 6 days a week for 6 months. They conclude that “in none of the patients […] an explicit homonymous change of the absolute field defects border [was] observed after training” (p. 30). Raemaekers et al. (38) perimetrically identified increased visual fields after RT but found that corresponding fMRI data “[...] could not account for the large increases in visual field size that were observed in some patients” (p. 872). Frolov et al. (39) attest that there are “remaining nagging questions as to the validity of [published data] and the clinical benefit” (p. 40), and thus the effectiveness of RT remains unclear (40).

The first main reason for skepticism regarding the efficacy of RT is the imprecision of existing diagnostic instruments used in previous studies to assess the extent and potential amelioration of visual field loss. Automated static perimetry, like the Humphrey® Field Analyzer (HFA)–the gold standard in clinical visual field assessment–do not offer sufficient accuracy–due to the limited number of displayed stimuli positions. In addition, these devices do not have an automatically continuous and stringent eye fixation control to clearly exclude eye movements for compensation of visual field deficits. Although automatic static perimetry systems sometimes offer some sort of technical control mechanism (e.g., blind spot stimulation) or visual control of eye fixation for the examiner, there is no reliable control for every (small) eye movement, (short) fixation loss or quick saccadic search behavior. In clinical practice, these inaccuracies might be acceptable for the benefit of a quick diagnosis. However, in a scientific context where the central question is about improvements after a neuropsychological intervention in the range of a single-digit degree of visual field, such inaccuracies stemming from compensating eye movements lead to false impressions of a larger visual field. This is especially true if a therapy elicits high hopes in the patients and trains them to pay attention to peripheral stimuli. In this context, another significant technical issue is that the HFA uses the so-called SITA (Swedish interactive thresholding algorithm) to shorten the diagnosis in clinical routine. This means that based on the patient’s age and neighboring test points, this algorithm estimates - based on a database - the luminance threshold of every test point [e.g., (41)]. This represents a sensible method in everyday clinical practice, but a source of potential biases in scientific studies where the focus is on the most exact assessment of potential improvements after a neuropsychological intervention.

Goldmann Perimetry (GP) offers no remedy either as GP is characterized by low retest reliability, as even the inventor of the device himself - Hans Goldmann - states that “[...] perimetry, and in particular kinetic perimetry, is an art. If one lets several young assistants examine the same patient [...] one will be astonished, even shocked about the difference in the results. It needs a long period of training until the results of two clinicians are comparable. “[Goldmann quoted after (42), p. 3]. Even the application of alternative, non-standard diagnostic instruments such as “microperimetry” could not resolve the controversy. To illustrate, Marshall et al. (26) and Reinhard et al. (37) both used microperimetry to examine the effects of RT. While Marshall et al. (26) found “modest but real expansions in visual fields” (p. 1027), Reinhard et al. (37), in contrast, observed–“in none of the patients” - a “change of the absolute field defect [...] after training” (p. 30). Against this background, Frolov et al. (39) emphasized that “the assessment of any potential visual restoration technique ultimately must rely on a reproducible and appropriate perimetric method.” (p. 36). Summarized, the main problem is the lack of a precise and reliable perimetric method that adapts for patients’ compensatory eye movements, particularly rapid (unconscious) saccades.

The second main reason for skepticism regarding the efficacy of RT is the selection of patient samples and the empirical framework, especially in larger studies. For example, Mueller et al. (29) included 302 patients with lesions in post-chiasmatic and/or pre-chiasmatic pathways due to stroke, trauma, tumor, or anterior ischemic optic neuropathy. These inclusion criteria allowed for a wide range of comorbid medical conditions (e.g., neglect). Similarly, Romano et al. (31) included 161 patients with homonymous visual field defects after post-chiasmatic insult, but provided no further differentiation of affected brain regions. Smaller studies–such as from Poggel et al. (30) with 9 patients–also show heterogeneous patient samples. Eight patients were diagnosed with postchiasmatic lesions leading to homonymous hemianopia, but one patient had optic nerve damage after a tumor surgery leading to bilateral heteronymous loss of vision. Gall et al. (43) included 85 patients with significantly differing etiology, ranging from ischemic infarction, traumatic brain injury, hemorrhagic infarctions, encephalitis, anterior ischemic optic neuropathy and arteritic optic nerve infarction. Additionally, 69 of 85 patients paid privately for participating in the training leading to authors’ and patients’ competing and personal interests. Summarized, a clear investigation of the efficacy of RT was obstructed by heterogeneous patient samples. As a result, comorbid disorders (e.g., neglect, prechiasmatic lesions) have confounded–in a considerable number of previous studies - an accurate analysis of the therapy’s potential effect on neuronal regeneration in early cortical areas.

The present study is the first of its kind to quantify the therapeutic effect of RT utilizing a specially developed and validated eye-tracking-based perimetric methodology that fully corrects for compensatory eye movements in real time (44). Also, it is the first study to include a highly selective sample of patients with visual field defects originating from lesions in early cortical areas, who conducted RT using VR goggles, which we specifically developed for this study (45).

(1) Acquired post-chiasmatic anopia (H53.4; ICD-10), (2) Lesioning of the occipital lobe, specifically primary visual cortex or optic radiations due to, e.g., stroke from thrombosis of the posterior cerebral artery (I63.33; ICD-10) or similar etiology [e.g., (7, 46)] (3) Perimetrically diagnosed “Chronic visual field defect” for a minimum of 3 months since clinical incident (7, 39, 47) (4) Older than 18 years.

(1) Patients with diagnosed cognitive deficits such as disorder of awareness, speech production, anosognosia, severe attentional problems, problems understanding and following instructions (e.g., R41.x, R44.x, F06.7; ICD-10) after significant temporal and/or parietal lobe damage (7) (2) Patients with unilateral visuospatial neglect (R29.5; ICD-10)

Patients were recruited via our partnering rehabilitation institution (Rehabilitation Center Grossgmain), training partner (Club Mobil) and with the help of press reports to the public. Information regarding our study was also disseminated by the Paris Lodron University Salzburg and the University Clinic Salzburg. Eligible patients were first interviewed in person or via telephone, comprehensive individual medical histories were obtained, and inclusion and exclusion criteria were assessed accordingly. These anamnestic data and neuropsychological examinations performed during previous hospitalization and rehabilitation were utilized to exclude confounding comorbidities and to ensure reliability and validity of visual field assessment and the self-performed execution of RT.

A total of 16 patients were included. This number is greater than in comparable studies who found significant improvements of stroke patients’ damaged visual field after RT [e.g., (22, 26, 30)]. Fourteen patients sustained cerebral infarction due to thrombosis of the posterior cerebral artery (I63.33; ICD-10). One patient additionally had middle cerebral artery ischemia (I63.33; ICD-10 and I63.51, ICD-10) and one patient was diagnosed with basal ganglion hemorrhage (I61.3, ICD-10).

The study was approved by the ethics committee of the University of Salzburg (Reference No. 39/2018) and guided by the fundamental principles of respect for the individual, the right to self-determination and informed decisions, i.e., informed consent, as noted in the Declaration of Helsinki. The study was registered in the ICMJE-approved registry German Clinical Trials Register (DRKS00025205).

The study qualifies as a clinical pre-post intervention study with 16 patients diagnosed with quadrantanopia or hemianopia after cortical lesions in early visual areas (see Table 1 for details). After an initial assessment - consisting of visual field diagnostics with a Humphrey Field Analyzer (HFA) and Goldmann Perimetry (GP)–the patients’ visual border area between intact and defect visual field was precisely assessed with the Eye Tracking Based Visual Field Analysis [EFA; (44)]. We focused on the inner 10 to 15° for three main reasons. (1) Central foveal areas are represented in the early visual cortex (V1) with a significantly larger area of cortical surface than more peripheral, outer areas (48). Following the logic of RT, the higher the number of neurons, the higher the probability of new connections between them. (2) By limiting diagnostic effort on these inner areas, we kept visual field assessments short and prevented erroneous behavioral responses due to cognitive overload, stress or fatigue. (3) Previous studies showed that improvements after RT ranged around 5 to 10 degrees of visual angle. Improvements of this size would be especially desirable in central areas of the visual field.

Results from the EFA provided the basis for individual rehabilitation configuration of the Salzburg Visual Field Trainer [SVFT; (45)] for every patient. Patients were extensively educated on the usage of the SVFT and the functionality of RT and instructed to exercise for 5 months, 6 times a week, 2 times a day for 30 min (SVFT was configured to automatically end training after 30 min). Training regime and sample size was based on previous studies reporting amelioration after RT [e.g., (22, 26, 29, 49)]. After around 2 months, the patients’ visual field was perimetrically reassessed using the EFA in order to adapt the training stimuli in the SVFT to potential changes in visual border characteristics.

The aim of the experiment was twofold: First, we investigated whether the visual field of stroke patients objectively increases after rehabilitation with RT. Second, a single question was used to determine whether the patients would experience/perceive a subjective increase of their intact visual field. The question was: “How do you rate the current status of your visual field?.” The potential answer ranged from “very bad” to “very good” and was provided on a horizontally aligned Visual Analog Scale (VAS). The methodology of VAS is a validated psychometric approach and widely used for subjective measurement of emotions, sensations, and other subjective feelings [e.g., (50)].

In an initial assessment session, the visual field of the included patients were measured by a trained ophthalmologist using conventional methods [Humphrey Field Analyzer (HFA) and Goldmann Perimetry (GP)]. This gave us an initial insight into the perimetric status of the patients and enabled configuration for the next diagnostic step. In this second assessment session and based on the results from GP, the inner visual field area (10–15 degrees of visual angle) of the patients’ respective border area were assessed in high resolution (1 stimulus = 1 angular degree) using our Eye Tracking Based Visual Field Analysis [EFA; (44)]. Figure 1 shows exemplary patient’s data and the logic of translation from GP to EFA.

The EFA is a computerized visual field test based on the principles of classical automated static perimetry. The special feature of the EFA, however, is the continuous eye fixation control, which (1) checks throughout whether the patient is fixating centrally and (2) compensates any deviation in real time by adapting the variance to the position of the currently presented test stimulus. Specific care was taken to ensure that the individual partial assessment with the EFA did not last longer than 15 min. Depending on the size of the visual subareas to be examined, breaks were scheduled to avoid generating assessment errors due to overexertion. In addition to automated real-time adjustment of test stimuli depending on patient eye movements, the EFA also controls the timing of responses to the presented stimuli. The presentation of the stimuli is varied using a randomized time interval. For example, if patients systematically press the response button - even though no stimulus was presented - the first three times the patient is instructed with a text not to do so. In addition, this behavior is counted automatically, even if no warning is displayed on the screen after the third time. For detailed information on the EFA see Leitner et al. (44).

Besides usage for pre-post assessment and investigation for potential changes after rehabilitation with RT, perimetric EFA results were also used for the individual placement of training stimuli (1 stimulus = 3 angular degrees) on the patients’ visual border area in the Salzburg Visual Field Trainer (SVFT). The SVFT is a virtual reality device–based on Google cardboard systems–for which we developed training software based on the principle of RT. For detailed information on the SVFT see Leitner et al. (45). Once the SVFT training goggles were configured to each patient’s individual border area location and extent, patients were given detailed instructions and education about the training regime, the logic behind RT, and the SVFT itself. We paid special attention to the fact that the patient should maintain a continuous fixation on the central fixation cross during the training. We repeatedly drew the patient’s attention to the fact that this must be always complied with. Every 2 months, an intermediate perimetric assessment was performed and–if necessary–the localization of the training stimuli in the SVFT were adjusted accordingly in case shifts in the visual border area were found. After completion of the intervention a final visual field assessment was performed. The described intervention procedure was not changed throughout the duration of the experiment. However, due to the COVID-19 pandemic some interim assessment dates had to be postponed. Consequently, the time between some intermediate assessments were shorter or longer than 2 months (also see the Results section and the Supplementary material). Figure 2 shows a summary of the experimental procedure.

Perimetric results from the EFA were classified as follows (based on (44)): (1) Perceived stimuli, (2) partially perceived stimuli and (3) unperceived stimuli. If the patient reacted to a presented stimulus, it was documented as perceived. If a stimulus was not perceived it was repeated after one randomized intermediate stimulus (which was excluded from analysis). If the stimulus was not perceived the first time but perceived at repetition, it was classified as a partially perceived stimulus. If the stimulus was not perceived both times, the stimulus was defined as unperceived. EFA plots illustrate perceived stimuli in green, partially perceived stimuli in brown and unperceived stimuli in red. For data analysis, perceived stimuli were calculated with 1 point, partially perceived stimuli with 0.5 points, and unperceived stimuli with 0 points.

In every first EFA assessment, a tight clustering of a generous number of stimuli was used to assess the exact border between defect and intact visual field as approximated from data of HFA and GP. Thereby, in later sessions, the number of test stimuli could be reduced in areas known to be clearly intact or defect. In order to ensure comparability between assessment dates, only test stimuli that were utilized in all assessment sessions were included for data analysis. This ensured no under-or overestimation of potential visual field changes.

The closer a visual field defect lies foveally, the more distressing for the affected patient. Especially in central areas where visual acuity is high, even a few angular degrees of improvement would be crucial for the patient. For this reason, we additionally calculated patients’ perimetric results specifically within 0° to 5° by weighting the decrease in acuity into analysis. This means that the closer a test stimulus was displayed foveally, the higher it was weighted in the analyzes [e.g., (51)]. Accordingly, test stimuli from 0° to 5° were weighted with the following values (in brackets): 0° (1), 1° (0.697), 2° (0.535), 3° (0.434), 4° (0.365), 5° (0.315).

Each end of the VAS scale regarding the subjective perception of the patients’ subjective status of his/her own visual field loss was defined by contrasting terms ranging from “very bad” (0%) to “very good” (100%). The position of the marking on the VAS made by the patient was measured, converted into the corresponding percentage value, and statistically analyzed.

On average, we perimetrically examined all patients for 3.1 times (SD = 0.6), with a mean of around 154 days (SD = 66.9) between first and last assessment date. Analysis of rehabilitation documentation from the SVFT records show that patients performed on average 1.4 (SD = 0.4) training sessions per day (also see Table 2).

Perimetric control values of the EFA did not show any conspicuous accumulations regarding “trigger happiness” in any patient. The number of mean false responses was 0.9 (SD = 1.1; MIN = 0; MAX = 5) in all perimetric assessment sessions.

Evaluation of subjective self-assessment regarding the size of intact visual field shows that patients reported a mean percentile improvement of 11.5%–changing from 51.0% (SD = 21.2) before RT to 62.4% (SD = 17.9) after completing RT (see left panel of Figure 3). Kolmogorov–Smirnov test (KS) indicates normal distribution and paired samples t-test shows a statistically significant difference and a medium to large effect size between first and last assessment date; t(15) = −2.356, 95% CI [−0.215, −0.011], p = 0.032, r = 0.52.

Analysis of patients’ visual field tests shows that mean stimulus detection rate before RT was 50.7% (SD = 10.6) and after RT 49.6% (SD = 13.1; see middle panel of Figure 3). KS indicates normal distribution and paired samples t-tests show no statistically significant differences between perimetric pre-post assessment; t(15) = 0.590, 95% CI [−0.288, 0.051], p = 0.564 (For details on the individual patients see the Supplementary material).

Note that only patients with defects in the inner 5° qualified for the following analysis (N = 12): Eccentricity weighted analysis of potential changes near the macula before and after RT shows that the calculated mean stimulus detection value before RT was 2.01 (SD = 0.34) and after RT 2.03 (SD = 0.32) (see right panel of Figure 3). KS indicates normal distribution and paired samples t-tests show no statistically significant differences between perimetric pre-post assessment; t(11) = −0.152, 95% CI [−0.280, 0.243], p = 0.882.