Introduction

Intellectual and developmental disabilities (IDD) encompass a spectrum of neurodevelopmental disorders that emerge during developmental phases, hallmarked by intellectual impairments and the inability to attain key developmental milestones1,2,3,4. These disorders frequently present comorbidly with behavioral conditions, including autism spectrum disorder (ASD) and cerebral palsy (CP), and exert significant societal, economic, and emotional impacts5,6,7,8. With a prevalence rate of 1–3% and an upward trend correlating with societal progress9,10,11, comprehensive assessments and tailored interventions are essential for optimal development.

Despite the importance of early identification and intervention for individuals with IDD12, current clinical practice relies heavily on parental questionnaires and expert clinical assessment rather than direct child observation13. Although advancements in gene microarray analysis and imaging studies14,15,16, the complex etiology and heterogeneous clinical manifestations of IDD make early and differential diagnosis from typical development (TD) challenging17,18,19.

Studies affirm the link between early infant gaze and communication and social abilities, with eye-gaze avoidance, a common feature in IDD, potentially affecting social engagement20,21,22,23. Eye movement analysis, reflecting brain function and behavior, is further emphasized by the intricate communication between nearly all cerebral regions and the ocular motor control system24,25,26,27,28. Benjamin et al.’s study supports this, indicating eye behavior as a reflection of attention deficits in fragile X syndrome, ASD, and TD individuals29. Thus, eye-movement patterns and visual attention may serve as promising biomarkers for cognitive and developmental disparities, aiding diagnosis and treatment of various diseases30,31,32,33,34.

Eye-tracking technology, including EOG, eye-tracking glasses, and remote eye trackers, is typically used to capture eye movement patterns such as saccades and fixations35,36,37,38. Among them, EOG, a non-invasive method quantifying electrical potential changes between the cornea and retina from eyeball movement, has been a key tool since the 1970s39,40,41. Its applications extend to diagnosing retinal diseases and facilitating EOG-based eye-writing communication42,43,44. Consequently, this study proposes EOG’s critical role in capturing visual attention allocation dynamics, reflecting attention processes in real-world scenarios. Nonetheless, the automatic, rapid, and non-repetitive nature of this phenomenon may hinder comprehensive EOG signal (EOGs) analysis45,46.

Our team has previously developed the Attention-related Visual Evocation System to address the challenge of EOG signal analysis. This system elicits attentional activation in children while simultaneously recording their EOGs during horizontal smooth pursuit with eyes viewing. The aim was to investigate attention-related phenomena in a controlled standard setting47. Given the suggested correlation between eye movement patterns and EOG waveforms48, we utilized the template matching technique to capture eye movements. The method’s efficacy was tested on a cohort of 4 TD individuals and 4 individuals with ASD, showing reliability and initial indications of promise in identifying specific subgroups within the IDD population49.

This study aimed to quantify attention in order to elucidate significant attentional disparities in individuals with IDD. We employed the Attention-related Visual Evocation System for precise EOG recordings and applied nonlinear dynamics methodologies for feature extraction, forming the EOG Signal Analysis (EOG-SA) method. We expanded our child participant sample and compared visual attention patterns in children with IDD and a matched control group using controllable standard visual stimuli, then contrasted these results with conventional assessment scales. We also explored various scenarios to understand real-life visual attention difficulties. This comprehensive approach underscored the potential of EOG-SA markers as a significant biobehavioral measure for IDD, promising to advance early detection, risk assessment, and personalized medicine progression (Fig. 1).

Figure 1
figure 1

Conceptual Framework of the Study. The images were created by Shi-yi Qi, one of the co-authors as the copyright holder. A schematic representation delineating the methodological approach and investigative workflow employed in the study, encapsulating the essential phases of investigation, data analysis procedures, and overarching research objectives.

Full size image

Methods

Participants

This investigation presents a multicenter case–control study. Participants were recruited from the Affiliated People’s Hospital of Fujian University of Traditional Chinese Medicine, Fujian Maternity and Child Health Hospital, and the Third People’s Hospital Affiliated to Fujian University of Traditional Chinese Medicine. The study protocol was approved by the Clinical Research Ethics Committees of the participating hospitals (approval numbers: 2021-016-02, 2021YJ042, 2020-kl-042). Registration of the study with the Chinese Clinical Trial Registry was completed, bearing the identifier ChiCTR-TRC-2200058387, and the initial trial registration date was scrupulously logged as 08/04/2022. Adhering to the stringent ethical standards of research, all study procedures were conducted in strict compliance with the updated Declaration of Helsinki, and we affirm that all methods were carried out in accordance with the relevant guidelines and regulations.

Participants in the IDD group met DSM-5 diagnostic criteria2, were aged 3–9, of either gender, and had guardian consent. Exclusion criteria included epilepsy, severe mental disorders, significant cerebral involvement (confirmed by CT or MRI), ocular surgery or disease history, and current medication use. The control group (TD) comprised age and sex-matched, clinically healthy children with typical development.

Child neurodevelopment assessment

The Infants-Junior Middle School Students’ Social-Life Abilities Scales (S-M scale), adapted from the Japanese S-M social life ability test50, was used to assess developmental and social maturity in children aged 6 months to 14 years. It covers six domains: self-dependence, locomotion, work skills, communication, socialization, and self-management51,52. For children under 6, the Gesell Developmental Schedule (GDS) was used for subgroup analysis. The GDS, a global child development scale developed by Yale University53, assesses neurological and intellectual growth in infants aged 1 month to 6 years across five domains: gross motor, fine motor, linguistic, adaptive, and personal-social skills. Results were expressed as a developmental quotient (DQ)54.

Apparatus

EOGs collecting procedure, attention-related visual evocation system

Following our previous work47,55, we devised an Attention-related Visual Evocation System to evoke attentional responses. This system included a display screen, visual stimuli, and a dynamic interface (Fig. 2A). Participants, either independently or with their parents, were positioned approximately 0.5 m from the screen to view video sequences. The procedure began with a cartoon image moving leftward and rightward in each 6-s cycle. The image was replaced in each cycle, with the entire process lasting for 60 s (Fig. 2 B/C). The system aims to provide a reliable tool for investigating attention-related phenomena in a controlled standard setting.

Figure 2
figure 2

Experimental Apparatus and Setup Configuration. The images were created by Shi-yi Qi, one of the co-authors as the copyright holder. A comprehensive visual array depicting the experimental tools and environment utilized in the study. (A) Electrooculography (EOG) acquisition system for monitoring eye movements. (B) User interface of the Attention-related Visual Evocation System, designed for stimulus presentation. (C) Sample video sequence as displayed by the Attention-related Visual Evocation System. (D) Diagrammatic representation detailing the specific placements of electrodes for EOG data collection. (E) Illustrative overview of the three experimental scenarios employed.

Full size image

We also developed an EOGs collection system utilizing a 24-bit A/D conversion chip ADS1299 and STM32F407 control chip. This system incorporated three additional electrodes for EOG collection, designed to study smooth pursuit eye movements. The horizontal EOGs were recorded by placing two electrodes at the outer canthus of the eyes and a reference electrode on the forehead (Fig. 2D). The impedance of all electrodes was maintained below 5 kΩ and the sampling was programmed using LabVIEW 7.1 (National Instruments, Austin, TX USA), with a sampling rate of 1000 Hz.

Evaluate the discrimination degree of application for EOG-SA across three different scenarios

To assess the applicability of EOG-SA across varied scenarios, we conducted three daily behavior tests (Fig. 2E) representing standard, guided, and freeplay scenarios. Each scenario represents a different meaning, with the standard scenario requiring participants to follow a predetermined eye movement trajectory, the guide scenario involving social interaction, and the freeplay scenario involving an object of interest. The specific set-up for each test was as follows:

Standard scenario Participants watched video sequences from the Attention-related Visual Evocation System for a duration of 5 min.

Guide scenario Participants were instructed by trained researchers to engage in a 5-min reading session with a children’s book.

Freeplay scenario Participants engaged in unstructured play with toys within the testing room for 5 min.

To ensure reliability, pediatricians were extensively trained and certified at the Affiliated People’s Hospital of Fujian University of Traditional Chinese Medicine, and regular quality-control examinations were conducted.

Ethics declarations

This investigation presents a multicenter case–control study. Participants were recruited from the Affiliated People’s Hospital of Fujian University of Traditional Chinese Medicine, Fujian Maternity and Child Health Hospital, and the Third People’s Hospital Affiliated to Fujian University of Traditional Chinese Medicine. The study protocol was approved by the Clinical Research Ethics Committees of the participating hospitals (Approval Numbers: 2021-016-02, 2021YJ042, 2020-kl-042). Registration of the study with the Chinese Clinical Trial Registry was completed, bearing the identifier ChiCTR-TRC-2200058387, and the initial trial registration date was scrupulously logged as 08/04/2022. Adhering to the stringent ethical standards of research, all study procedures were conducted in strict compliance with the updated Declaration of Helsinki, and we affirm that all methods were carried out in accordance with the relevant guidelines and regulations.

Data preprocessing, and feature extraction

Data preprocessing

EOGs span an amplitude range of 50–3500 µV and a frequency range of 0.1–10 MHz56,57. As in previous work47, preprocessing of raw data was required to eliminate noise, using a 6-order Butterworth filter as a 50 Hz notch filter and a Chebyshev low-pass filter to remove frequencies above 10 Hz. Further analysis was performed utilizing Matlab (Matlab R2022a, The MathWorks Inc., Natick, Massachusetts). Unconscious blinking, an inevitable artifact during EOG recording, was eliminated using the Moving-Average Method, which calculates the mean and standard deviation of the instantaneous spike rate within a moving interval58. It is capable of effectively removing jitter, resulting in smoother EOGs that facilitate the identification of targeted eye movements. A sliding window with length of 500 sampling points and a step size of 1 sampling point was employed47.

Feature extraction

Dynamic time warping (DTW) for EOG waveforms

In this study, we combined human-labeling and computer vision techniques to monitor attentional state and calibrate eye movements, using a hybrid approach. Initially, eye movement data (EOGs) representing the subject’s attentional state were generated, and subsequently used as templates in Dynamic Time Warping (DTW) to measure similarity with test EOGs47. This process involved expert clinicians assessing attentional state during template construction. Attention-related Visual Evocation System was used to engage participants in observing video sequences while clinicians recorded and marked time intervals. Three marker intervals were necessary, as illustrated in Fig. 2B. If test EOG waveforms resembled template waveforms within a 6-s timeframe, they were identified as target EOGs for that interval, indicating focused attention.

The DTW algorithm calculated the similarity metric, functioning as a distance measure capable of handling non-linear distortion and phase errors59,60,61. This algorithm uses a dynamic programming approach to determine minimal distance between two time series, improving waveform alignments62. Typically, the distances between data points reflect similarity, with closer points indicating higher similarity63. The computation of DTW distance is carried out as follows: the proximity of data points is indicative of their level of similarity, the template and test EOGs are denoted by M = (m1, m2, m3 . . . mn) and T = (t1, t2, t3 . . . tq), respectively, where n and q represent the length of the signals. Firstly, the matrix D with i rows and j columns is established as the sequence distance matrix, where D (i, j) = (mi − tj)base. Generally, the squared Euclidean distance is used as the base distance are the Euclidean distances from the point to point of the M and T sequences in the corresponding rows and columns. This approach facilitates the generation of a cost matrix C (i, j), wherein C (1,1) corresponds to D (1,1), and C-Values are calculated using Eq. (1):

$$C\left( {i,j} \right) = D\left( {i,j} \right) + min\left\{ {\begin{array}{*{20}l} {C\left( {i,j - 1} \right)} \hfill \\ {C\left( {i - 1,j} \right)} \hfill \\ {C\left( {i - 1,j - 1} \right)} \hfill \\ \end{array} } \right.$$
(1)

The computation of distances between M and T yielded C (n, q).

To enhance identification and evaluation robustness, we refined the Template threshold determination using the mentioned techniques. Clinicians annotated three to five attentional time intervals, indicating the corresponding system boxes. The template threshold test was performed thrice, each lasting 60 s. Subsequently, we calculated the DTW distances between the template and EOG waveforms within these intervals, using the average distance as the Template threshold.

Furthermore, our system interface provided real-time DTW values every 6 s, which we will refer to as RT-DTW in the following sections (Fig. 2B).

Approximate entropy (ApEn) for EOG signals

Introduced by Pincus in 199164, Approximate entropy (ApEn) is a nonlinear statistic quantifying regularity and complexity in time series data, particularly suitable for short, noisy series65,66. Higher ApEn values represent increased fluctuations or irregularities, while lower values signify reduced fluctuation67. This methodology has been successfully applied to physiological time series analysis68,69. The formula for ApEn is defined as follows:

$${\text{ApEn}}\left( {m,r} \right)\mathop {\lim }\limits_{{{\text{n}} \to \infty }} \left[ {{\upvarphi }^{m} \left( r \right) - \varphi^{m + 1} \left( r \right)} \right]$$
(2)

In our study, we acquired 5000 longitudinal data sets and calculated a single ApEn value (m = 2, R = 0.2sd) for every 1000 data points using the “pracma” command in R programming language (version 3.6.3). The average of the five computed results was then determined.

Correlation Dimension (D2) for EOG signals

The correlation dimension (D2), another method to measure time series complexity, was introduced by Grassberger and Procaccia in 198370. It characterizes chaotic attractors by evaluating their estimated geometric structure71.

Consider \(X_{n}\), a vector of length m, given by:

$$X_{n} \left( {m,\tau } \right) = \left[ {X_{n} ,X_{n + \tau } , \ldots ,X_{{n + \left( {m - 1} \right)\tau }} } \right],n = 1,2, \ldots ,N_{m}$$
(3)

Among these variables, m represents the embedding dimension, τ denotes the time delay, and \(N_{m}\) signifies the dimension of the phase-space.

Let \(C\left( r \right)\), , referred to as the correlation integral, be ascertained as follows:

$$C\left( r \right) = \mathop {\lim }\limits_{N \to \infty } \frac{1}{{N\left( {N - 1} \right)}}\mathop \sum \limits_{i \ne j} \theta \left( {r - \left| {X_{i} - X_{j} } \right|} \right)$$
(4)

The D2 can be defined as the limit value:

$$D2\left( m \right) = \mathop {\lim }\limits_{r \to 0} \left[ {\frac{lnC\left( r \right)}{{lnr}}} \right]$$
(5)

In this study, we quantified D2 values by determining the regression curve slope using Matlab.

To sum up, the signals extracted subsequent to the aforementioned process, referred to as Electrooculogram Signal Analysis (EOG-SA) in this context, will be subjected to an exploration of their correlation with disease scales.

Statistical analysis

The statistical analysis was conducted using SPSS (version 22 for Windows, SPSS Inc., USA). Continuous variables were presented as mean ± standard deviation (SD), while categorical variables were detailed as frequencies (percentages). Upon confirming the non-normal distribution of the data through normality tests, we employed Spearman coefficients for correlation assessments and utilized either the Independent t-test or Mann–Whitney rank-sum test for comparisons between two groups. For comparisons involving more than two groups, Kruskal Wallis with Dunn’s multiple comparison post-test was utilized. Linear regression analysis was applied for continuous outcome variables, while logistic regression was used to model associations with Visual Attention Patterns for binary outcome variables. The ROC curve was employed for further validation. All tests were two-tailed, with P < 0.05 denoting statistical significance.

Results

Demographic and clinical characteristics of participants

From January 2021 to March 2023, we recruited participants from respected medical institutions including the Affiliated People’s Hospital of Fujian University of Traditional Chinese Medicine, Fujian Maternity and Child Health Hospital, and the Third People’s Hospital Affiliated to Fujian University of Traditional Chinese Medicine. Of the 140 enrolled families, 13 were excluded due to data loss, comorbid conditions, or other reasons, leaving 127 participants. The study was successfully completed by 63 patients with IDD and 64 healthy adolescents, all of whom were right-handed Asians. Parents provided informed consent and no adverse events were reported. Details on participant selection and demographics are illustrated in Fig. 3 and Table 1, respectively.

Figure 3
figure 3

Clinical Trial Progression Flowchart. This flow diagram delineates the structured progression of the clinical trial, mapping out the sequential stages from participant enrollment through to the completion of the study.

Full size image
Table 1 Demographic and clinical characteristics of participants.
Full size table

Comparative analysis of EOG signal analysis (EOG-SA) in two distinct developed profile groups of children

In our study, we extract features from EOGs recorded in the Attention-related Visual Evocation System, hereby denoted as EOG Signal Analysis (EOG-SA), specifically referring to the template threshold, ApEn, and D2 values. As shown in Table 2, the IDD group displayed significantly higher template threshold and D2 values than the TD group (P < 0.001), but the ApEn values did not significantly differ between the groups (P > 0.05).

Table 2 Differences in Template threshold, ApEn and D2 values between the two groups.
Full size table

Examination of the relationship between scale and EOG-SA

We further probed the association between scales and EOG-SA, encompassing comparative analysis across different scales in two groups, and the relationship of EOG-SA with developmental levels and disease severity.

The IDD group exhibited significantly higher template threshold and D2 values on both the S-M scale and GDS compared to the TD group (P < 0.001), but not ApEn (P > 0.05) (Fig. 4 A/D). Moreover, significant associations were observed between template threshold/D2 values and individual items and total score of the scales (P < 0.001), while ApEn values correlated with S-M scale items, excluding total score and work skills (P > 0.05), but not with GDS (P > 0.05) (Fig. 4 B/E).

Figure 4
figure 4

Correlating Intellectual and Developmental Disabilities (IDD) with Electrooculogram Signal Analysis (EOG-SA) Metrics. A comprehensive analysis illustrating the association between IDD and various EOG-SA parameters. (A) Illustrates significant differences in template threshold, approximate entropy (ApEn), and correlation dimension (D2) values between control and IDD groups, as gauged by the total S-M scale score (***P < 0.001). (B) Depicts the results from Spearman’s correlation analysis linking template threshold, ApEn, and D2 values with the S-M scale. (C) Presents a comparative analysis of the aforementioned EOG-SA parameters across different severity grades of IDD, as delineated by the S-M scale score (**P < 0.01, ***P < 0.001 versus borderline; #P < 0.05 versus mild). (D) Highlights differences in EOG-SA metrics between groups based on the total GDS scale score (***P < 0.001). (E) Shows Spearman’s correlation analysis outcomes for template threshold, ApEn, and D2 values in relation to the GDS. (F) Compares EOG-SA parameters across IDD severity grades as defined by the GDS score, indicating significant variations (***P < 0.001 versus borderline; ##P < 0.01 versus mild or moderate).

Full size image

Finally, upon stratifying disease severity according to total S-M or GDS scale score (Table 3), we examined EOG-SA’s correlation with disease severity in TD/DD individuals, aiming to identify visual attention patterns associated with varying developmental delay degrees. Severity of the condition was determined using the total S-M score72,73, with our data indicating a negative correlation between severity scores on the S-M scale and both thresholds and D2 (r =  − 0.77, P < 0.001; r =  − 0.63, P < 0.001), whereas ApEn did not show significant correlation (P > 0.05). Notably, individuals with moderate and severe or profound IDD displayed significantly higher template thresholds than those with borderline IDD (P < 0.001). However, no significant differences were found in template thresholds between patients with borderline, mild, and moderate IDD (P > 0.05), nor in terms of ApEn and D2 (P > 0.05) (Fig. 4C). Furthermor, when severity was classified based on the total GDS score54,74, a clear negative correlation was found between severity scores on the GDS scale and both thresholds and D2 (r =  − 0.81 P < 0.001; r =  − 0.71 P < 0.001). Patients with severe or profound IDD had significantly higher template thresholds compared to those with borderline or mild or moderate IDD (P < 0.001, P < 0.01). Yet, no significant differences were observed in other indicators (P > 0.05) (Fig. 4F).

Table 3 Distribution of severity grade.
Full size table

Linear regression analyses

We conducted a series of linear regression analyses to evaluate the influence of participant age, gender, and developmental outcomes on EOG-SA. Age, gender, or each subscale of the S-M scale or GDS were used as independent variables, with EOG-SA as the dependent variable. The results revealed a significant effect of age on ApEn (β = − 0.001, P = 0.02). Furthermore, personal-social skills significantly impacted template threshold (β = − 0.81, P = 0.03), while both gross motor and personal-social skills influenced D2 significantly (β = − 0.03, P = 0.01; β = − 0.83, P = 0.02). These positive findings were visually represented using scatter plots and regression lines (Fig. 5A). Further details are provided in Table 4.

Figure 5
figure 5

Evaluating Electrooculogram Signal Analysis (EOG-SA) Across Diverse Scenarios Using Linear Regression and Discrimination Metrics. (A) Depicts the outcomes of linear regression analysis applied to EOG-SA data. (B) Provides a comparative assessment of real-time Dynamic Time Warping (DTW), approximate entropy (ApEn), and correlation dimension (D2) values across three distinct scenarios, revealing significant disparities (***P < 0.001 versus standard scenario; ###P < 0.001, ##P < 0.01, comparisons between TD individuals and those with IDD).

Full size image
Table 4 Results of linear regression analysis.
Full size table

Analysis results on discrimination degree of EOG-SA application across scenarios

Our subsequent analysis sought to discern potential variations in EOG-SA application across different groups and scenarios. The IDD group showed significantly higher real-time DTW(RT-DTW) and D2 values than the TD group across scenarios (###P < 0.001, ##P < 0.01), but no significant ApEn differences were observed (P > 0.05). Both TD and IDD groups displayed significantly increased RT-DTW and ApEn in guide and freeplay scenarios compared to the standard scenario (P < 0.001), with no significant D2 variation (P > 0.05) (Fig. 5B).

Results of binary logistic regression analysis and ROC curve evaluation

Finally, binary logistic regression analysis was employed to examine the relationship between EOG-SA and IDD, followed by ROC curve analysis for further evaluation. Independent variables included EOG-SA measures (template threshold, ApEn, and D2 values), while the dependent variable was the presence or absence of IDD, diagnosed using the S-M scale or GDS.

Binary logistic regression results, based on the S-M scale, indicated significant influences of both template threshold and D2 values on the presence of IDD (β = 0.06, P = 0.000; β = 1.12, P = 0.001), with ApEn showing no effect (β = 2.30, P = 0.25). Furthermore, ROC curve analysis revealed AUCs of 0.91 and 0.85 for the template threshold and D2, respectively (Fig. 6A). Using the Youden method, threshold values for template threshold and D2 were determined as 79 and 2.19, respectively, with sensitivities of 74.2% and 90.3%, and specificities of 95.0% and 63.3%.

Figure 6
figure 6

Performance Analysis of Disease Prediction Models Using Electrooculogram Signal Analysis (EOG-SA). This figure presents the analytical performance of EOG-SA-derived metrics in disease prediction models. Receiver Operating Characteristic (ROC) curve analyses are illustrated, with template threshold depicted in blue and Correlation Dimension (D2) in red, across both the S-M scale and Global Deterioration Scale (GDS). The diagnostic sensitivity and specificity are represented by paired columns in shades of blue and purple, respectively, for Dynamic Time Warping (DTW) and D2 within these assessment scales. A. Within the S-M scale context, DTW achieved a sensitivity of 74.2% and a specificity of 95.0%, whereas D2 reached a sensitivity of 90.3% and a specificity of 63.3%. B. Within the GDS framework, DTW’s sensitivity and specificity stood at 77.6% and 93.5%, respectively, while D2 displayed sensitivity and specificity of 95.9% and 71.0%, correspondingly.

Full size image

Further analysis based on GDS revealed significant effects of both template threshold and D2 values on the presence of IDD (β = 0.06, P = 0.002; β = 1.64, P = 0.003), with no effect from ApEn (β = 2.19, P = 0.45). ROC curve analysis yielded AUCs of 0.91 for both measures, with thresholds of 79 and 2.19, respectively. Sensitivity and specificity at these thresholds were 77.6% and 95.9% for template threshold, and 93.5% and 71.0% for D2 (Fig. 6B).

Discussion

This study evaluated cognitive and developmental quotient using the S-M scale, and GDS for certain age-specific participants. The findings revealed higher template threshold and D2 values in individuals with IDD than those with TD, indicating excellent discriminant validity. Both values correlated well with both scales, demonstrating robust criterion validity. The template threshold showed a stronger association with all parameters than D2 and could discriminate the severity of IDD. Despite no statistical significance between borderline and mild IDD, similar trends were observed across both scales, possibly due to the limited sample size. Consequently, the internal consistency of the template threshold was deemed acceptable, affirming its exceptional properties.

The linear regression results showed a significant effect of personal-social skills on template threshold, suggesting that higher personal-social skills correlate with lower template thresholds. This aligns with prior research linking visual patterns to social disability27,75. Similar effects were observed for D2, with both gross motor and personal-social skills significantly impacting. Gross motor variables unexpectedly emerged from the analyses, providing external validity to our findings, and supporting previous studies linking gross motor development to cognition76,77 and development78,79. Scholars have found that patients with developmental dyscalculia exhibit deficits in visual perception and motor integration, as well as mildly delayed gross motor milestones80,81. Current research focuses on links between visual characteristics, social and cognitive deficits, and developmental disorders in IDD patients. Our study highlights the potential value of EOG analysis with D2 for assessing gross motor development. The lack of a relationship in the S-M scale remains unclear, possibly due to the age specificity of the GDS, used for children under six82. This age-related difference may be more pronounced at younger ages, a view supported by our findings with ApEn. Future age-stratified studies are planned to test this hypothesis.

Furthermore, to integrate eye-movement and attentional state recognition into social interactions, we evaluated the performance of EOG-SA in three distinct scenarios. Each task, from tracking a pre-defined eye movement trajectory in the standard scenario, engaging in social interaction in the guide scenario, to focusing on an object within an unbounded setting in the freeplay scenario, conveyed a unique connotation.

Here, employing real-time DTW (RT-DTW) calculations at 6-s intervals, we observed distinct visual attention patterns between individuals with IDD and TD across various scenarios, demonstrating the efficacy of RT-DTW and D2 in discerning individuals with TD from those with IDD. Despite higher RT-DTW and ApEn values in the guide and freeplay scenarios compared to the standard,, suggesting that these scenarios require more attention and cognitive effort, no significant differences were noted between the guide and freeplay scenarios. This may be due to the heightened variability of eye movements in real-world settings, whereas the templates utilized in our analysis were derived from the standard scenario, thereby limiting their ability to differentiate between these scenarios. Our results underscore the need for individualized attention pattern consideration in designing interventions for IDD. While D2 analysis effectively differentiated between IDD and TD, it failed to distinguish between scenarios. Interestingly, ApEn analysis, though less effective in disease classification, successfully distinguished between scenarios While no significant differences were detected between the guide and freeplay scenarios, a slight inclination towards higher values in the freeplay scenarios was observed in both groups.

In IDD clinical trials, precise discernment of attentional alterations is crucial for early disease diagnosis and effective monitoring83,84. Hence, enhancing the sensitivity and objectivity of outcome measures has been a paramount pursuit for researchers85,86.

Eye movement patterns encompass the minute, involuntary ocular motions that transpire during diverse visual tasks. Traditional devices, such as eye-tracking glasses or remote eye trackers, employ infrared light to meticulously track the eyes’ position and record their intricate movements87,88,89. These investigations commonly assess the duration, frequency, and dwell time of gaze on different elements of a visual scene, including facial features and areas of interest (AOIs)90,91,92,93. While these conventional methods provide valuable insights into information processing94,95,96, they are not typically utilized in social contexts97. Given the complex and dynamic nature of visual attention98, relying solely on fixation patterns and absolute percent fixation times may fail to capture subtle eye movement patterns, ultimately leading to misleading conclusions that lack sensitivity and objectivity99. Thus, while eye-tracking studies are promising, their outcomes often apply more to groups than individuals100,101, offering limited clinical insights.

EOG, with its high signal-to-noise ratio, objectivity, sensitivity, and ease of recording, is valuable in situations where conventional eye-tracking devices may be impractical102,103,104. It allows for the evaluation of time-dependent visual attention dynamics and the transformation of eye-tracking metrics into practical biomarkersrather than mere proportional distribution measures100. Expanding this research to encompass natural social interactions and settings is crucial, as it can provide quantifiable indicators of cognitive processes and attentional focus associated with eye movements105, complementing psychophysical and neural cognition measurements (Fig. 7).

Figure 7
figure 7

Comparative Analysis of Measured Parameters in Electrooculography and Eye-Tracking Technologies. The schematic delineates a side-by-side comparison of the principal parameters measured by electrooculography (EOG) and eye-tracking technologies, illustrating the distinct performance characteristics and methodological differences of each technique in monitoring ocular movements.

Full size image

Finally, to further validate the robustness of our study, specifically the assessment of the template threshold and D2 as valuable methods for distinguishing between TD and IDD, we employed Binary Logistic Regression Analysis Results to prevent overfitting. The classifier modeling performance was appraised via the ROC curve. The AUCs of the template threshold were 0.91 on both scales, whereas the AUCs of D2 were 0.85 (S-M scale) and 0.91 (GDS), respectively, yielding the largest Youden indexes and optimal cut-off points of the elastic strain ratio (79 for template threshold and 2.19 for D2). These cutoffs were evaluated utilizing the S-M scale for the former and the GDS for the latter, resulting in a sensitivity of 74.2% and 77.6% for the template threshold, and a specificity of 95.00% and 93.5%, respectively. As for D2, the sensitivities were 90.3% and 95.9%, and the specificities were 63.3% and 71.0%. The high specificity rates of template threshold indicate low false-positive rates106, thereby enabling accurate identification of IDD107 and reducing the likelihood of misdiagnosis among participants108. Therefore, individual EOG waveforms in the attentional state serve as a template for individual diagnoses, which can be defined as a description of the observed behavior of visual attention. This approach emphasizes behavioral analysis at the individual level, rather than comparing differences at the population level.

Naturally, it is common for certain diagnostic methods to exhibit specificity but lack sensitivity, resulting in the possibility of missed diagnoses109,110. This shortfall can be mitigated by utilizing D2. Sensitivity, which measures the capacity to accurately identify true positives, becomes crucial in this context111. A high sensitivity value for D2 implies a lower probability of missed diagnoses112, thus minimizing the misclassification of individuals with IDD as TD. The template threshold and D2 exhibit high specificity and sensitivity, respectively, in IDD diagnosis, effectively curtailing both misdiagnoses and missed diagnoses. Consequently, these parameters could serve as potential diagnostic markers for IDD, with thresholds set at 79 (template threshold) and 2.19 (D2).

Performance disparities observed may be due to the distinct meanings conveyed by the data. To pinpoint individual attentional change patterns, we utilized a template matching algorithm based on DTW47. Customizing the template threshold to each individual was crucial for state recognition. Furthermore, ApEn, a measure of information production rate113, denotes the randomness and irregularity within an interactive ensemble114, while D2, a complexity metric, represents the strange attractor’s magnitude in the phase space and characterizes an individual’s overall state115. ApEn and D2, thus, reflect different physiological aspects. While ApEn characterizes dynamic changes during the process of information integration, D2 is more apt for distinguishing between health and disease states. These findings concur with previous EMG116 or EEG117,118 studies.

Limitations

Despite the encouraging nature of the findings presented in this study, it is imperative to acknowledge the limitations that warrant attention. The primary limitation pertains to the absence of a formal and direct measurement of the subjects’ IQ, with information on the degree of IDD derived solely from scales and clinical notes. A second limitation concerns the lack of stratification in the study group, such as age stratification, and the absence of an assessment of diagnoses such as ASD. Furthermore, a third limitation is attributed to the relatively abbreviated duration of the three scenarios, which imposes constraints on the comprehensive interpretation of the results. Nonetheless, in light of these limitations, we contend that the EOG-SA presented in this study hold the potential to function as an early biomarker of IDD.

Conclusions

Our study highlights the efficacy of integrating EOG, Attention-related Visual Evocation System, and nonlinear analysis techniques (EOG-SA) to discern Visual Attention Patterns between individuals with IDD and their TD counterparts. The template threshold based on DTW demonstrated superior discrimination and criterion validity, effectively distinguishing the severity of the condition. Notably, the template threshold and D2 demonstrated high specificity and sensitivity, respectively, for the diagnosis of IDD, thereby mitigating the risk of misdiagnosis and missed diagnosis. These tools also show potential for assessing personal-social skills and gross motor development. Meanwhile, ApEn effectively differentiates diverse application scenarios. This individualized matching technology, which requires no frequent calibration or prior training, is particularly suitable for the IDD population, including young children. It accurately monitors attention over time and identifies pattern changes. This study is the most comprehensive application of EOG in IDD to date, with the EOG-SA presented offering a quantifiable measure to detect subtle changes in visual attention patterns, aiding early diagnosis and disease monitoring.