The physical part of the system (hardware) includes (Table 1): (a) a three-dimensional display screen configured to show stereoscopic anaglyphs previously stored in a database; (b) a chin rest upon which the subject's chin is placed, positioned 40 cm from the viewing screen; (c) red–green 3D glasses; (d) an upper camera that measures the distance between the tip of the index finger and the monitor using computer vision technology; (e) a computer with the processing unit configured to manage the equipment software; and (f) a second monitor that can be viewed by the clinician (Fig. 1). Devices a, b, and d are incorporated into a compact system with custom housing and its own lighting system (Fig. 2).

Fig. 1.

Layout of the physical components of the system.

(0.16MB).

Fig. 2.

Prototype of the compact system.

(0.12MB).

Additionally, the software comprises:

Graphic design and professional 3D animation software were used to model, texture and generate multimedia content for three original characters, developed to be attractive to children: parrot-pirate, fairy and robot (Fig. 3).

Fig. 3.

Three-dimensional models.

(0.07MB).

These are 3D models of complex figures that are located in the most distal plane and have a reference point (guide point) that the subject must point to with the tip of their index finger. The character and guide point are displayed so that there are no monocular cues that are helpful for the user to determine the distance between the objects. Each of these guide points appears at a different distance: further away in the parrot-pirate and closer in the robot (Fig. 4). This way, it will be easier to detect stereopsis with the parrot-pirate and more difficult with the robot, being the fairy the intermediate difficulty.

Fig. 4.

Each character has an object with a reference guide point that protrudes from the distal plane where it is located (pirate's parrot, tip of the fairy's wand and one of the robot's arms).

(0.09MB).

The guide point (parrot's beak in the pirate model, tip of the magic wand in the fairy model, and end of the fist in the robot model), is located (1) in the monitor centre, and (2) shifted to one side respect the main object, leaving the character displaced to one side; thus preventing monocular cues of superposition.

Using the known measurements of the design of the different images (IPD and distance of the figures on the screen (δ)), the exact point where subjects should place their fingertip for detection by the camera was calculated mathematically in milimeters (Fig. 5), i.e., the geometric prediction.

Fig. 5.

Geometric prediction. Distance “d” depends on the interpupillary distance (IPD) of each patient, and the distance of the figures on the screen (δ).

(0.02MB).

The models are shown with a background image that allows clear separation between the background and the 3D model. To that end, a background with a white noise random-dots pattern was used, as shown in Fig. 6.

Fig. 6.

Images of each of the stereoscopic models: (a) stereoscopic model of a fairy with a magic wand with a star at the tip, leaning forward; (b) stereoscopic model of a pirate with a parrot flying ahead at a certain distance; (c) stereoscopic model of a robot with the left fist outstretched.

(0.25MB).

The sample size was calculated in order to detect significant differences between different stereopsis grades, assuming an α error of 5 % and a ß error of 10 %. Based on these calculations, the number of subjects needed was at least 80.

The study included 94 healthy volunteers who met the following inclusion criteria: (1) age range: 4–65 years; (2) corrected visual acuity 20/20 or higher; (3) absence of far and near tropia by cover test; (4) correct fusion with Worth light test; (5) measurable stereopsis on the Randot Stereostest; (6) absence of past or current eye or systemic diseases.

In the study, a descriptive analysis of the demographic and clinical variables of the subjects was carried out. Quantitative variables are summarized by mean, median, and standard deviation. Categorical variables are reported as frequencies and percentages.

Subjects were examine and included by ophthalmologist investigators. Each of the participating subjects underwent an anamnesis, including their age and sex, and an ophthalmological examination that included autorefractometry (including IPD), visual acuity, biomicroscopy, fundoscopy, cover test, Worth test, and eye dominance. The subjects’ depth perception was then analyzed using the DALE3D platform.

In addition, a stereopsis study was performed using the following standard stereoscopic tests: Lang I and II (Vision Assessment Corporation, Illinois, USA), TNO (TNO test for stereoscopic vision, 5th edition, Lameris Ootech, Nieuwegein, Netherlands) and Randot (Cicle test, Stereo Optical Company, Inc., Chicago, Illinois). The Randot circles test was selected for the DALE3D correlation study because it offers the most levels of stereoacuity.

To validate the test, the repeatability and reliability of the test were calculated, analyzing the data using R software (version 4.0.2).

To assess possible differences between different groups of the study sample (sex, age, spherical equivalent, IPD), subgroup analyses was performed. After verifying that the variables did not conform to normal using the Kolmogorov–Smirnov test, non-parametric tests were used. The Mann–Whitney test was used to compare two groups and the Wilcoxon test was used to compare more than two groups.

Each subject was given a satisfaction survey for the different stereopsis tests (TNO, Randot, Lang and DALE3D) with a score of 1(worst) to 5 (best) represented by emoticons. Three question about each test were asked: Did you like the test? Did you like the images shown? Was it difficult to understand?

The studies of this diagnostic platform were carried out in accordance with the Declaration of Helsinki and approved by the Clinical Research Ethics Committee of the Hospital General de Elda (Alicante) / FISABIO (Foundation for the Promotion of Biomedical Health Research of the Valencian Community). All participating subjects were informed about the study and signed written informed consent to participate; confidentiality and data protection was guaranteed in accordance with Spanish data protection law (Organic Law 7/2021).

The study included 94 subjects, 25 children and 69 adults, with a mean age of 34.2 ± 18.9 years; 36.2 % were men and 63.8 % were women. Two-thirds of participants (60.6 %) showed right ocular dominance compared to 39.4 % with left ocular dominance. The spherical equivalent of both eyes showed a myopic tendency (OD −0.17 ± 2.4; OS −0.01 ± 2.4). The mean IPD recorded was 61.3 ± 4.67 mm (Table 2).

The mean results of the standard stereoscopic tests are shown in arcsec, obtaining a maximum for stereoacuity with the Randot test (60.32 arcsec) and a minimum with the Lang I test (550 arcsec). With respect to the DALE3D platform, depth perception is shown in mm, the maximum being in pirate model (54.86 mm) and the minimum in robot (12.82 mm) (Table 3).

The DALE3D platform obtained very good repeatability with an ICC (Table 4) close to 1 in all models (0.94–0.87), and CV (Table 5) between 0 % and 5 % in all models (0.1–0.26). It is represented by Bland Altman plots for each model in Fig. 7.

Fig. 7.

Bland–Altman plots for each of the models.

(0.11MB).

No statistically significant differences were observed between the two age groups: children (0–18 years) and adults (>18 years) (p value > 0.05), either in the standard stereoscopic tests or in the DALE3D test (Table 11).

Among the groups established according to the spherical equivalent (hyperopia and emetropia, and myopia), no statistically significant differences (p value > 0.05) were observed in any of the stereoscopic tests (Table 12).

The analysis of depth sensation according to sex (male or female) showed statistically significant differences (p value < 0.05) in all models, with higher measurements in females than in males (Table 13).

In terms of the correlation between the IPD and the depth sensation detected, no statistically significant differences were obtained in any of the models (p value > 0.05) (Table 14).

The test that was most liked was the DALE3D (90 % of the maximum score) and the least liked was the Lang test. The favourite models were the DALE3D (88 % maximum score) and the least popular were those of the Lang test. The test reported as the easiest to perform by the participants was the DALE3D (73 % very easy) and the most difficult to perform was the TNO test (Fig. 8).

Fig. 8.

Satisfaction survey graphs.

(0.17MB).