ASD early screening system based on refractive state and eye biological parameter measurement

Abstract

The invention relates to the technical field of autism screening, in particular to an ASD early screening system based on refractive state and eye biological parameter measurement, refractive parameters and eye biological parameters of a target child are collected through high-precision ophthalmology equipment, the refractive parameters comprise a spherical equivalent value SE and a cylindrical power CYL, and the refractive parameters comprise a spherical equivalent value SE and a cylindrical power CYL; the eye biological parameters comprise the eye axial length AL, the corneal curvature K1 / K2, the anterior chamber depth ACD and the axial length / corneal curvature ratio AL / CR. Through joint modeling of four parameters of SE, CYL, AL / CR and ACD and multi-parameter collaborative diagnosis, AUC can be effectively improved, the defect that the AUC numerical value of a traditional single-parameter method is low is overcome, and sensitivity and specificity are effectively improved; the judgment threshold is automatically adjusted according to age stratification (4-6 years old, 7-9 years old and 10-12 years old), the screening accuracy of the young children is improved, and the dynamic threshold is optimized so as to adapt to different crowds.

Description

Technical Field

[0001] The present invention relates to the technical field of autism screening, and particularly to an early screening system for ASD based on the measurement of refractive status and eye biological parameters. Background Art

[0002] Autism Spectrum Disorder (ASD) is a complex neurodevelopmental disorder. Its early screening relies on behavioral observation and scale assessment (such as ADOS, ADI-R), which has problems such as strong subjectivity, long time consumption, and easy missed diagnosis. Existing research shows that there are significant differences in the ocular biological parameters (such as refractive status, anterior chamber depth) between children with ASD and normal children. However, traditional ophthalmological research only focuses on refractive correction and does not deeply explore its association with neurodevelopment. In recent years, although machine learning technology has been applied to the behavioral data analysis of ASD, it has not combined objective ocular biological parameters to construct a prediction model. The main defects of the existing technology include:

[0003] (1) Traditional screening relies on behavioral characteristics, cannot quantify the differences in biological parameters, and lacks objective biological markers;

[0004] (2) Existing research mostly analyzes refraction or axial length in isolation, ignoring the synergistic effect of multiple parameters, which will lead to the limitations of single-parameter analysis;

[0005] (3) Ophthalmic devices are only used for data collection and are not deeply integrated with intelligent algorithms to achieve real-time analysis, resulting in the separation of devices and algorithms.

[0006] Therefore, it is urgent to design an early screening system for ASD based on the measurement of refractive status and eye biological parameters to solve the above problems. Summary of the Invention

[0007] The purpose of the present invention is to provide an early screening system for ASD based on the measurement of refractive status and eye biological parameters to solve the above deficiencies in the existing technology.

[0008] To achieve the above purpose, the present invention provides the following technical solutions:

[0009] An early screening system for ASD based on the measurement of refractive status and eye biological parameters, comprising the following steps:

[0010] (a) Collect the refractive parameters and eye biological parameters of the target child through a high-precision ophthalmic device, where the refractive parameters include spherical equivalent value SE and cylinder power CYL, and the eye biological parameters include axial length AL, corneal curvature K1 / K2, anterior chamber depth ACD, and axial length / corneal curvature ratio AL / CR;

[0011] (b) Standardize the collected parameters, and the calculation formula is:

[0012] X' = (X - μ) / σ;

[0013] Where X is the original parameter value, μ is the mean of the corresponding parameter, and σ is the standard deviation;

[0014] (c) Input the standardized parameters into the pre-trained random forest model, and the feature weight function of the model is:

[0015] S = 0.255·SE + 0.253·ACD + 0.249·CYL + 0.243·AL / CR;

[0016] (d) Compare the risk score S output by the model with the preset threshold. When S ≥ 0.728, it is determined as high risk of ASD;

[0017] (e) Output a visual diagnostic report including the ROC curve (AUC = 0.741) and the confusion matrix.

[0018] Preferably, it further includes a longitudinal monitoring module, which can record the temporal changes of parameters and calculate the developmental deviation index: as shown in the appendix Figure 2

[0019] When Δ > 2.58, an alarm is triggered.

[0020] Preferably, the range of the spherical equivalent value SE in the refractive parameters is from -5.00DS to +5.00DS, and the range of the cylinder power CYL is from -4.00DC to +4.00DC.

[0021] Preferably, the standardization process includes imputing missing data using the k-nearest neighbor algorithm (k = 5) and performing 3σ truncation processing on outliers.

[0022] Preferably, the construction of the random forest model includes: setting the number of trees to 500, the maximum depth to 8, the minimum number of samples per node to 5, and using the Gini coefficient as the splitting criterion.

[0023] Preferably, it further includes a data calibration module. When using the IOLMaster device, it is necessary to calibrate it through a standard model eye first, and the calibration error is controlled within ±0.02mm (AL) and ±0.25D (K value); the eye biometric parameter acquisition device includes an integrated device of an autorefractor, an optical biometer, and a corneal topographer.

[0024] Preferably, the visual diagnostic report generation module is integrated with a dynamic threshold adjustment function, which can automatically optimize the determination threshold according to age stratification (4 - 6 years old, 7 - 9 years old, 10 - 12 years old).

[0025] Preferably, it further includes a data security module, which encrypts patient information using the AES-256 algorithm, and the storage period of biometric data does not exceed 72 hours after the diagnosis is completed.

[0026] Preferably, the ROC curve is generated by the bootstrap resampling method (n = 1000 times), and the confidence interval is calculated by the DeLong test.

[0027] Preferably, when it is determined as high risk, the system automatically generates an intervention guide including a refractive correction plan (hyperopia compensation formula: ADD = 0.75 × SE) and sensory integration training suggestions.

[0028] In the above technical solution, an ASD early screening system based on the measurement of refractive status and eye biological parameters provided by the present invention: (1) By jointly modeling with four parameters of SE, CYL, AL / CR, and ACD, and multi-parameter collaborative diagnosis, the AUC can be effectively improved, changing the drawback of the low AUC value of the traditional single-parameter method, and effectively improving the sensitivity and specificity; (2) Automatically adjust the determination threshold according to age stratification (4 - 6 years old, 7 - 9 years old, 10 - 12 years old), improve the screening accuracy of young children, and optimize the dynamic threshold to adapt to different populations; (3) With the provided real-time intervention support, the system automatically generates a personalized intervention plan (such as the hyperopia compensation formula ADD = 0.75 × SE) after determining high risk, shortening the diagnostic intervention cycle to within 24 hours, and improving the diagnostic efficiency. Description of the Drawings

[0029] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those of ordinary skill in the art, other drawings can also be obtained based on these drawings.

[0030] Figure 1 It is a step schematic diagram provided by an embodiment of an ASD early screening system based on the measurement of refractive status and eye biological parameters of the present invention.

[0031] Figure 2 It is a schematic diagram of the formula for calculating the developmental deviation index provided by an embodiment of an ASD early screening system based on the measurement of refractive status and eye biological parameters of the present invention.

[0032] Figure 3 It is a schematic diagram of the standardization processing formula provided by an embodiment of an ASD early screening system based on the measurement of refractive status and eye biological parameters of the present invention.

[0033] Figure 4 It is a schematic diagram of the demographic distribution of the ASD group and the control group provided by an embodiment of an ASD early screening system based on the measurement of refractive status and eye biological parameters of the present invention (a. The violin plot shows the age data of the two groups; b. The bar chart shows the gender data of the two groups).

[0034] Figure 5 Schematic diagram of the refractive states of the ASD group and the control group provided for an embodiment of the ASD early screening system based on the measurement of refractive states and eye biological parameters of the present invention (the violin plots show the spherical degree (SPH) data of the two groups; the violin plots show the cylinder degree (CYL) data of the two groups; the violin plots show the spherical equivalent (SE) data of the two groups).

[0035] Figure 6 Schematic diagram of the ocular biometric parameters of the ASD group and the control group provided for an embodiment of the ASD early screening system based on the measurement of refractive states and eye biological parameters of the present invention

[0036] Figure 7 Schematic diagram of the correlation analysis between the spherical equivalent (SE) and the ocular biometric parameters in the ASD group provided for an embodiment of the ASD early screening system based on the measurement of refractive states and eye biological parameters of the present invention Figure 8 Schematic diagram for the correlation analysis of the refractive states and ocular biometric parameters of children with ASD Detailed implementation manners

[0037] In order to enable those skilled in the art to better understand the technical solutions of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings.

[0038] As Figure 1-7 shown, an ASD early screening system based on the measurement of refractive states and eye biological parameters provided by an embodiment of the present invention, an ASD early screening system based on the measurement of refractive states and eye biological parameters, includes the following steps:

[0039] (a) Collect the refractive parameters and eye biological parameters of the target child through a high-precision ophthalmic device, where the refractive parameters include the spherical equivalent SE and the cylinder degree CYL, and the eye biological parameters include the axial length AL, corneal curvature K1 / K2, anterior chamber depth ACD, and axial length / corneal curvature ratio AL / CR;

[0040] (b) Perform standardization processing on the collected parameters, and the calculation formula is:

[0041] X′ = (X - μ) / σ;

[0042] where X is the original parameter value, μ is the mean of the corresponding parameter, and σ is the standard deviation;

[0043] (c) Input the standardized parameters into a pre-trained random forest model, and the feature weight function of the model is:

[0044] S = 0.255·SE + 0.253·ACD + 0.249·CYL + 0.243·AL / CR

[0045] (d) Compare the risk score S output by the model with a preset threshold. When S ≥ 0.728, it is determined as a high risk of ASD;

[0046] (e) Output a visual diagnostic report including the ROC curve (AUC = 0.741) and the confusion matrix.

[0047] Preferably, it further includes a longitudinal monitoring module, which can record the temporal changes of parameters and calculate the developmental deviation index: as shown in the appendix Figure 2

[0048] When Δ > 2.58, a warning is triggered.

[0049] Preferably, the range of the spherical equivalent value SE in the refractive parameters is from -5.00DS to +5.00DS, and the range of the cylinder power CYL is from -4.00DC to +4.00DC.

[0050] Preferably, the standardization process includes imputing missing data using the k-nearest neighbor algorithm (k = 5) and performing 3σ truncation processing on outliers.

[0051] Preferably, the construction of the random forest model includes: setting the number of trees to 500, the maximum depth to 8, the minimum number of samples per node to 5, and using the Gini coefficient as the splitting criterion.

[0052] Preferably, it further includes a data calibration module. When using the IOLMaster device, it needs to be calibrated first through a standard model eye, and the calibration error is controlled within ±0.02mm (AL) and ±0.25D (K value); the eye biometric parameter acquisition device includes an integrated device of an auto-refractor, an optical biometer, and a corneal topographer.

[0053] Preferably, the visual diagnostic report generation module is integrated with a dynamic threshold adjustment function, which can automatically optimize the determination threshold according to age stratification (4 - 6 years old, 7 - 9 years old, 10 - 12 years old).

[0054] Preferably, it further includes a data security module, which encrypts patient information using the AES-256 algorithm, and the storage period of biometric data does not exceed 72 hours after the diagnosis is completed.

[0055] Preferably, the generation of the ROC curve adopts the bootstrap resampling method (n = 1000 times), and the confidence interval calculation adopts the Delong test.

[0056] Preferably, when it is determined to be a high risk, the system automatically generates an intervention guide including a refractive correction plan (hyperopia compensation formula: ADD = 0.75 × SE) and sensory integration training suggestions.

[0057] Example 1

[0058] Step 1: Data collection and calibration

[0059] Use an integrated device (autorefractor + IOLMaster 700) to collect the refractive parameters (SE, CYL) and eye biometric parameters (AL, K1 / K2, ACD) of the target child. The device needs to be calibrated through a standard model eye to ensure that the AL measurement error ≤ 0.02 mm and the K value error ≤ 0.25 D.

[0060] For children who do not cooperate, use the fast scan mode (≤ 0.4 seconds / time), and collect 3 times continuously and take the median value.

[0061] Step 2: Data standardization and processing

[0062] Standardize the original data according to the formula: as shown in the appendix Figure 3 shown;

[0063] where μ and σ are calculated based on the historical database (e.g., μ of SE = +0.12 D, σ = 1.35 D);

[0064] For missing data, use k-NN imputation (k = 5, distance metric uses Manhattan distance), and outliers are truncated with 3σ.

[0065] Step 3: Random forest model construction

[0066] The training data contains 191 samples (95 ASD + 96 controls), and the input features are SE, CYL, AL / CR, ACD;

[0067] Model parameters: number of trees = 500, maximum depth = 8, minimum number of samples per node = 5, Gini coefficient splitting criterion;

[0068] Feature weight assignment is calculated through Gini importance, and the final model outputs a risk score:

[0069] S = 0.255·SE + 0.253·ACD + 0.249·CYL + 0.243·AL / CR

[0070] Step 4: Risk assessment and output

[0071] Input the standardized parameters into the model. If S ≥ 0.728, it is determined to be a high risk (the threshold is determined by maximizing the Youden index);

[0072] The system generates a visual report, including the ROC curve (Delong test 95% CI: 66.94% - 81.19%), the confusion matrix, and intervention suggestions;

[0073] For high-risk cases, start the longitudinal monitoring module and calculate the developmental deviation index according to the formula: as shown in the Figure 2 attachment;

[0074] When Δ > 2.58, trigger an alarm (corresponding to p < 0.01).

[0075] Step 5: Hardware system integration

[0076] Device side: The optometer and biometer are connected to the main control computer through the PCIe interface, and the sampling rate ≥ 10Hz;

[0077] Software side: Deploy the model using the Python 3.8 + PyTorch framework, and the visual interface supports real-time parameter curve display.

[0078] Example 2

[0079] Take a 5-year-old boy as an example:

[0080] Sampling parameters: SE = +1.25D, CYL = -1.00D, AL = 22.1mm, ACD = 2.85mm, AL / CR = 2.89;

[0081] After standardization, input into the model, and calculate the score S = 0.255×1.25 + 0.253×2.85 + 0.249×

[0082] (-1.00) + 0.243×2.89 = 0.801;

[0083] Judge that S > 0.728 is high risk, and the system recommends refractive correction (ADD = 0.75×1.25 = +0.94D) and sensory integration training;

[0084] After 6 months of reexamination, Δ = 3.21 (SE change +0.50D, AL increase 0.3mm), trigger an alarm and recommend neurodevelopmental assessment.

[0085] Example 3

[0086] Autism spectrum disorder (ASD) is a complex neurodevelopmental disorder characterized by persistent impairments in social interaction and communication skills, as well as restricted and repetitive behaviors. These core features typically emerge in early childhood and affect multiple developmental domains such as cognitive, sensory, and motor functions. The prevalence of ASD is on the rise globally, with data from the United States indicating that 1 in 36 children is diagnosed with the disorder. A similar trend is observed in China, where some studies estimate the prevalence of ASD in children to be approximately 1%. Despite the complex etiology of ASD, involving genetic, epigenetic, and environmental factors, there is a consensus on the importance of early and accurate diagnosis. Early detection of ASD allows for evidence-based interventions that can significantly improve the developmental outcomes and quality of life of children with ASD and their families.

[0087] In addition to the challenges in behavior and social communication, children with ASD often exhibit unique sensory processing patterns that can affect visual function. Global studies consistently show a higher incidence of visual impairments in this group. Systematic reviews indicate that 22.5% - 44% of children with ASD have clinically significant refractive errors, with astigmatism being particularly more common than in neurotypical children, although the incidence of myopia and hyperopia is similar to that in the general child population. In China, a cross-sectional study in Beijing found that 45.1% of the eyes of children with ASD and intellectual disabilities had astigmatism, 47.2% had refractive errors, and 57.7% of the refractive errors were uncorrected. A study in Taiwan Province of China involving 3,551 children with ASD showed an increased risk of amblyopia, anisometropia, and strabismus, and a higher prevalence of hyperopia and astigmatism, although the incidence of myopia was similar to that in the general population. These findings suggest that refractive problems may be more prevalent in the ASD population compared to neurotypical children. Furthermore, abnormal refractive states may affect the developmental trajectory of children with ASD by influencing their ability to interact with educational materials and social environments, highlighting the need for systematic vision screening and personalized visual interventions for this vulnerable group.

[0088] In addition to refractive problems, children with ASD may exhibit unique ocular biometric characteristics such as AL, K, ACD, and LT. There are two major limitations in the current research field: firstly, traditional ASD research mainly relies on behavioral assessments and lacks quantitative analysis of ocular biological parameters; secondly, existing refractive development studies mainly target typically developing populations and overlook the ASD-specific refractive pathways and ocular biometrics. Importantly, abnormal refractive parameters may reduce the quality of visual input, thereby exacerbating the social avoidance behavior of children with ASD due to the sensorimotor integration process. The underlying pathological mechanism of this negative cycle remains unclear. Therefore, systematically studying the refractive development characteristics of children with ASD has important clinical value in the field of ophthalmology, contributing to the development of screening strategies for ASD and introducing a new research framework for understanding neurodevelopmental problems in ASD.

[0089] This study aims to explore the refractive development and ocular biometric characteristics of children with ASD. Using high-precision ocular biometric technology, key parameters such as corneal curvature and axial length are quantitatively measured. A connection model between refractive parameters and ASD clinical characteristics is established through machine learning algorithms to answer the following questions: (1) Is the refractive development pattern of children with ASD different from that of normally developing children? (2) Can specific refractive abnormalities (such as high astigmatism, anisometropia) serve as biomarkers for early identification of ASD? The research results are expected to drive important clinical progress, leading to the creation of ASD-specific visual screening protocols and improving visual perception quality through refractive correction. In addition, identifying specific biomarkers may improve ASD screening efforts;

[0090] Study subjects

[0091] Ninety-five children with ASD, aged between 4 and 12 years, were recruited from 14 special education centers in Handan between March and June 2024. The control group included 96 healthy children, matched by age and gender. All participants underwent a comprehensive eye examination, following the guidelines of the Declaration of Helsinki and obtaining ethical approval from Tianjin Eye Hospital (number: 2022009).

[0092] Study subjects

[0093] Ninety-five children with ASD, aged between 4 and 12 years, were recruited from 14 special education centers in Handan between March and June 2024. The control group included 96 healthy children, matched by age and gender. All participants underwent a comprehensive eye examination, following the guidelines of the Declaration of Helsinki and obtaining ethical approval from Tianjin Eye Hospital (number: 2022009). The study has been registered on ClinicalTrials.gov (registration number: NCT06122519). Inclusion criteria included: aged between 4 - 12 years, ASD diagnosis for more than one year, normal intraocular pressure, and no significant ocular pathology. Exclusion criteria included: congenital cataract, corneal disease, or a history of eye surgery.

[0094] Examination methods

[0095] Children with ASD and the normal control group received the same examinations. The anterior segment and fundus were examined using a handheld slit lamp and fundus camera. Refractive examinations included uncorrected distance visual acuity (UCDVA, using a numerical or HOTV visual acuity chart), spherical equivalent (by SPOT vision screening and retinoscopy), and ocular biometric parameters (such as AL, K1, K2, AST, AL / CR, ACD, CCT, WTW, LT, and VT). Children with ASD showed lower cooperation during the examinations, requiring more patience from the examiners and their special education teachers. All examiners were senior doctors and senior optometrists specializing in pediatric ophthalmology. The examiners received professional training and passed an assessment before participating in data collection and strictly followed the operating standards. The instruments were calibrated using a model eye before measurement, and data on refractive errors and ocular biometric parameters were collected and analyzed. Emmetropia was characterized by an SE between -0.50 DS and +0.50 DS. Astigmatism was characterized by an SE greater than 0.50 DS, and an SE greater than 2.00 DS indicated high astigmatism. SE (SE = S + C / 2) was recorded as the value of the refractive error.

[0096] Statistical analysis

[0097] Data analysis was performed using SPSS 27. For normally distributed data, the mean and standard deviation were reported, and an independent samples t-test was conducted. For non-normally distributed data, the median and interquartile range were reported, and the Wilcoxon rank-sum test was used. Logistic regression analysis for multivariate analysis was performed with a significance level of α = 0.05.

[0098] Random forest models, support vector machine models, KNN models, decision tree models, and naive Bayes models were developed to identify statistically significant ASD risk predictors. The evaluation effects of different models were compared, and a reasonable model was selected for machine learning, aiming to discover reliable ASD risk predictors;

[0099] Results

[0100] Demographic distribution

[0101] The study included 95 children with ASD (70.5% male, 29.5% female) and 96 control group children (68.8% male, 31.2% female), with median ages of 8 years and 7 years, respectively. There were no statistically significant differences between the two groups in terms of age or gender (age: z = -0.16, p = 0.873 > 0.05; gender: z = -0.588, p = 0.557 > 0.05). (See Figure 7 and 8 )

[0102] Changes in refractive status

[0103] The refractive status differences between the control group and the ASD group were statistically analyzed. The results showed significant differences in spherical power (SPH) between the two groups (ASD group: 0.25 ± 1.10 D, control group: -0.25 ± 2.10 D, z = -4.936, p < 0.01), significant differences in cylinder power (CYL) (ASD group: -0.75 ± 1.50 DC, control group: -0.50 ± 0.50 DC, z = -2.744, p = 0.006 < 0.01), and significant differences in spherical equivalent (SE) (ASD group: 0.00 ± 1.30 D, control group: -0.50 ± 1.80 D, z = -3.400, p = 0.001 < 0.01). ASD children showed a higher tendency for hyperopia and higher astigmatism values. To more clearly show the refractive status differences between the two groups, violin plots of SPH, CYL, and SE were drawn (see Figure 5 ).

[0104] Changes in ocular biometric parameters in ASD patients

[0105] A comparative analysis of the ocular biometric parameters of the ASD group and the control group was performed. The results showed statistically significant differences in anterior chamber depth (ACD) (t = 2.201, p = 0.029 < 0.05) and vitreous thickness (VT) (t = 2.056, p = 0.041 < 0.05) between the two groups. Compared with the control group, the ASD group had lower ACD and VT values. No statistically significant differences were found in other indicators (all p-values were greater than 0.05). (See Figure 7 )

[0106] Correlation between SE and ocular biometric parameters in the ASD group

[0107] A correlation analysis of the refractive status and ocular biometric parameters of ASD children was performed. The results showed that spherical equivalent (SE) was significantly negatively correlated with axial length (AL) (r = -0.377, p < 0.01) and significantly negatively correlated with the ratio of axial length to corneal curvature (AL / CR) (r = -0.359, p < 0.01). In addition, SE was significantly negatively correlated with anterior chamber depth (ACD) (r = -0.237, p < 0.05), significantly positively correlated with lens thickness (LT) (r = 0.266, p < 0.05), and significantly negatively correlated with vitreous thickness (VT) (r = -0.419, p < 0.01), but not significantly correlated with K1, K2, AST, WTW, and CCT (all p-values were greater than 0.05). (See Figure 8 )

[0108] Binary logistic regression analysis of refractive and biometric factors

[0109] Binary logistic regression analysis was performed on the refractive status and ocular biometric parameters of the ASD group and the control group. The results showed that the spherical equivalent (SE) (OR = 1.732, 95% CI: 1.245 - 2.410), cylinder power (CYL) (OR = 0.504, 95% CI: 0.346 - 0.734), axial length / corneal curvature ratio (AL / CR) (OR = 84.505, 95% CI: 1.188 - 6012.705), and anterior chamber depth (ACD) (OR =

[0110] 0.201, 95% CI: 0.043 - 0.930) were significantly associated with ASD (all p values were less than 0.05). Compared with the control group, children in the ASD group showed a higher tendency for hyperopia, higher astigmatism, lower axial length / corneal curvature ratio, and shallower anterior chamber depth.

[0111] When ROC curves of SE, CYL, AL / CR, and ACD were plotted (with specificity on the x-axis and sensitivity on the y-axis), the combined diagnostic method was more significant than single factors in predicting the risk of ASD, with an AUC value of 0.741 (95% CI: 66.94% - 81.19%). This means that AL / CR, ACD, and refractive status (SE, CYL) can be potential predictors of ASD, and combined evaluation can identify children with ASD with high sensitivity and specificity. These children should undergo scale assessment to confirm the diagnosis. Early detection of these ocular biometric parameters may enable more proactive clinical intervention, thus improving the visual and developmental outcomes of children at high risk or with diagnosed ASD. (The ROC curve is shown in Figure 6 A, and the summary of the ROC result AUC is shown in Figure 7 );

[0112] Figure 6 In: A. Plot the ROC curves of SE, CYL, AL / CR, and ACD;

[0113] A random forest model was developed to identify predictors of ASD. The weight results showed that SE and ACD were significant predictors, with weights of 25.50% and 25.28% respectively (where the x-axis represents weight / contribution and the y-axis represents predictors);

[0114] C. Confusion matrix of the random forest test set results;

[0115] D. Confusion matrix of the support vector machine test set results;

[0116] Machine learning model for evaluating ASD prediction

[0117] Develop random forest models, support vector machine models, KNN models, decision tree models, and naive Bayes models to identify statistically significant ASD risk predictors. Compare the evaluation effects of different models and select a reasonable model for machine learning. When evaluating the effectiveness of various machine learning models, the random forest model is the most stable. This model focuses on identifying predictors of ASD, especially SE, CYL, AL / CR, and ACD. Feature weights reveal the degree of contribution of each factor to the model, and the sum is 1. The results show that SE and ACD account for 25.50% and 25.28% of the model respectively, highlighting their key roles in model development. This indicates that SE and ACD are particularly valuable for early ASD screening and diagnosis. (The analysis results of the machine learning model are as Figure 6 shown).

[0118] Discussion

[0119] This study deeply explored the refractive status and ocular biometric characteristics of children with autism spectrum disorder (ASD), highlighting the significant differences from children in the control group. We evaluated refractive error and astigmatism in 95 children with ASD (95 eyes), using 96 neurotypical children (96 eyes) as the control group. The study results showed that the ASD group had a higher tendency for hyperopia, while the control group was more prone to myopia, and astigmatism was more significant in the ASD group. Scharre JE et al. found that 44% of children with ASD had refractive errors, among which astigmatism was 17.6%. Ikea et al. studied 154 children with ASD and found that 40% had eye diseases, among which refractive errors were the most common, followed by strabismus and amblyopia. Similarly, Khanna's study showed that the incidence of astigmatism was 26%, and 35% had significant refractive errors. In a study in Turkey, 22% of 324 children with ASD had significant refractive errors. These basic studies consistently showed that children with ASD had refractive errors and astigmatism to varying degrees, supporting the results of this study. However, the Qian team reported that the refractive status of children with ASD was normal, which may be due to differences in sample size, research methods, and selection criteria.

[0120] From birth to adolescence, the refractive state changes continuously as the eyes develop [^27-28^]. Studies by Haugen et al. have shown that preschool children (0-6 years old) mainly show hyperopia. As they grow older, the proportion of hyperopia gradually decreases, while the proportion of myopia gradually increases. Classic epidemiological evidence indicates that educational level plays a causal role in the development of myopia. The general conclusion of extensive research is that close reading and writing significantly increase the prevalence of myopia [^30-32^]. Children with ASD tend to have a higher rate of hyperopia and a lower rate of myopia, which may be related to the fact that children with ASD usually spend less time on reading and close writing tasks. The refractive state of ASD patients may be related to abnormal neurodevelopment, indicating that the refractive state can be used as a potential biomarker for ASD. High hyperopia or low myopia may be related to the unique visual information processing mode of ASD patients, providing clues for further research on the visual perception mechanism of ASD.

[0121] This study found that there was a significant correlation between the spherical equivalent (SE) and ocular biometric parameters in children with ASD: it was negatively correlated with axial length (AL), axial length / corneal curvature ratio (AL / CR), anterior chamber depth (ACD), and vitreous thickness (VT) (all p<0.05), while it was positively correlated with lens thickness (LT) (p<0.05). Notably, children with ASD showed lower ACD and VT values (p<0.05), while there were no significant differences in other parameters between the two groups (p>0.05). These results suggest that children with ASD with shorter axial length, lower axial length / corneal curvature ratio, shallower anterior chamber depth, increased lens thickness, or reduced vitreous thickness are more likely to have a hyperopic shift.

[0122] Currently, the predictive factors for ASD mainly include various psychological and behavioral characteristics, and the ocular predictive factors mainly focus on eye movement and visual attention [^33-36^]. There are relatively few studies on ocular biometric parameters. This study conducted a multivariate analysis of the refractive state and ocular biometric parameters in the ASD group and the control group. The results showed that the spherical equivalent (SE), cylinder power (CYL), axial length / corneal curvature ratio (AL / CR), and anterior chamber depth (ACD) were related to the risk of ASD onset. Compared with the control group, children in the ASD group showed a stronger tendency for hyperopia, higher astigmatism, lower axial length / corneal curvature ratio, and shallower anterior chamber depth.

[0123] By plotting the ROC curves of SE, CYL, AL / CR, and ACD, it was found that using these biological parameters in combination for prediction was more diagnostically significant than single factors. Research has shown that the axial length / corneal curvature ratio (AL / CR), anterior chamber depth (ACD), and refractive status (SE, CYL) can serve as potential predictors of ASD. Joint evaluation of these parameters can significantly improve the sensitivity and specificity of predicting the risk of ASD. Therefore, evaluating ocular biometric parameters helps identify children at high risk of ASD. These children should undergo further scale analysis and clinical symptom assessment. In cases where cooperation is limited, early identification of these ocular parameters may provide more opportunities for clinical intervention, thus improving the visual and developmental outcomes of children at high risk or with a confirmed diagnosis of ASD.

[0124] Machine learning model evaluation showed that the random forest model demonstrated superior stability in identifying ASD predictors. The model identified four key features: SE, CYL, AL / CR, and ACD. Feature importance analysis showed that SE (25.5%) and ACD (25.3%) were the main contributors, together accounting for more than 50% of the model's predictive ability. This highlights their potential clinical value as biomarkers for early ASD screening.

[0125] This cross-sectional study had dual methodological innovations in ASD research: (1) integrating ocular biometric and refractive features to identify neurodevelopmental biomarkers, different from traditional behavioral / neurological paradigms; (2) applying machine learning (random forest modeling) for the first time to quantify predictive ocular parameters. As the first study of children with ASD aged 4 - 12 years in the Handan region of China, this study filled a key evidence gap in the regional and global context. Comparative analysis identified four key biomarkers associated with the risk of ASD: SE, CYL, AL / CR, and ACD. Feature importance analysis showed that SE (25.5%) and ACD (25.3%) were the main predictors, together accounting for 50.8% of the model weight. The multi-parameter model showed clinically significant diagnostic accuracy (sensitivity 85.2%, specificity 91.7%), superior to single-parameter methods. This machine learning-based biomarker strategy can actively monitor high-risk populations and, through early targeted interventions, has the potential to improve visual and neurodevelopmental outcomes.

[0126] Despite these advances, there are still some limitations in this study. First, since children with ASD usually have difficulty expressing themselves and cooperating with examinations, we were unable to comprehensively collect factors such as genetic factors, visual environment, and behavior to deeply study the refractive development characteristics of children with ASD. Second, the study was limited to the Handan area of Hebei Province, where the economic and medical conditions are better than those in rural and mountainous areas, which may lead to selection bias. As a cross-sectional study, it cannot fully support the characteristics of refractive progression in children with ASD. Longitudinal studies can further explore the dynamic relationship between refractive status and ocular biometric parameters. In addition, due to the relatively homogeneous study group, the lack of comparison with the refractive characteristics of other types of special children may result in the biological indicators analyzed not fully representing the ocular characteristics of children with ASD. To better determine the correlation between ocular biometric indicators and ASD diagnosis and prediction, well-designed studies with larger sample sizes and exclusion of other influencing factors are needed in the future. Finally, this study did not perform cycloplegic refraction, which may lead to an overestimation of myopia and an underestimation of hyperopia. Although the study sample size was reasonable within the scope of the study, larger and more diverse samples will enhance the generalizability of the research results. In addition, although this study focused on ocular biometric data, it did not investigate how these factors affect the clinical manifestations of ASD or the potential effects of visual interventions.

[0127] Future research should aim to fill these gaps by studying the effects of refractive correction or other visual therapies on the developmental pathways of children with ASD. In addition, longitudinal studies can provide clearer insights into the temporal relationship between ocular biometric abnormalities and the emergence of ASD symptoms. Expanding the sample size is crucial for validating the consistency of the refractive status and ocular biometric characteristics of children with ASD and establishing a reference range matched to age and gender. Applying deep learning techniques can help analyze the complex relationships between refractive status, ocular biometric parameters, and ASD behavioral characteristics, potentially identifying subtype characteristics. Since refractive power (SE / CYL) and ocular structural parameters may change with age, it is necessary to conduct longitudinal studies to explore the temporal correlation between the ASD developmental process and visual system abnormalities. Studying the effects of refractive correction or visual training (such as binocular vision rehabilitation) on improving the core symptoms of ASD will also be valuable.

[0128] Only some exemplary embodiments of the present invention have been described above by way of illustration. Without doubt, for those of ordinary skill in the art, the described embodiments can be modified in various different ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and description are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. An early screening system for ASD based on the measurement of refractive state and eye biometric parameters, characterized in that, It includes the following steps: (a) Collect the refractive parameters and eye biometric parameters of the target child through a high-precision ophthalmic device, where the refractive parameters include the spherical equivalent value SE and the cylinder power CYL, and the eye biometric parameters include the axial length AL, corneal curvature K1 / K2, anterior chamber depth ACD, and axial length / corneal curvature ratio AL / CR; (b) Standardize the collected parameters, and the calculation formula is: X′ = (X - μ) / σ; where X is the original parameter value, μ is the mean value of the corresponding parameter, and σ is the standard deviation; (c) Input the standardized parameters into a pre-trained random forest model, and the feature weight function of the model is: S = 0.255·SE + 0.253·ACD + 0.249·CYL + 0.243·AL / CR; (d) Compare the risk score S output by the model with a preset threshold. When S ≥ 0.728, it is determined as a high risk of ASD; (e) Output a visual diagnostic report including the ROC curve (AUC = 0.741) and the confusion matrix.

2. The early screening system for ASD based on the measurement of refractive state and eye biological parameters according to claim 1, characterized in that It also includes a longitudinal monitoring module that can record the temporal changes of the parameters and calculate the development deviation index: When Δ > 2.58, an alarm is triggered.

3. An early screening system for ASD based on the measurement of refractive state and eye biological parameters according to claim 1, characterized in that, In the refractive parameters, the range of the spherical equivalent value SE is from -5.00DS to +5.00DS, and the range of the cylinder power CYL is from -4.00DC to +4.00DC.

4. An early screening system for ASD based on the measurement of refractive state and eye biological parameters according to claim 1, characterized in that, The standardization process includes imputing missing data using the k-nearest neighbor algorithm (k = 5) and performing 3σ truncation processing on outliers.

5. An early screening system for ASD based on the measurement of refractive state and eye biological parameters according to claim 1, characterized in that, The construction of the random forest model includes: setting the number of trees to 500, the maximum depth to 8, the minimum number of samples per node to 5, and using the Gini coefficient as the splitting criterion.

6. The early screening system for ASD based on refractive state and eye biometric parameter measurement according to claim 1, wherein It also includes a data calibration module. When using the IOLMaster device, it needs to be calibrated first through a standard model eye, and the calibration error is controlled within ±0.02mm (AL) and ±0.25D (K value); the eye biometric parameter collection device includes an integrated device of an autorefractor, an optical biometer, and a corneal topographer.

7. An early screening system for ASD based on the measurement of refractive state and eye biological parameters according to claim 1, characterized in that, The visual diagnostic report generation module is integrated with a dynamic threshold adjustment function, which can automatically optimize the determination threshold according to age stratification (4 - 6 years old, 7 - 9 years old, 10 - 12 years old).

8. An early screening system for ASD based on the measurement of refractive state and eye biological parameters according to claim 1, characterized in that, It also includes a data security module that encrypts patient information using the AES-256 algorithm, and the storage period of biometric data does not exceed 72 hours after the diagnosis is completed.

9. The early screening system for ASD based on the measurement of refractive state and eye biological parameters according to claim 1, characterized in that, The ROC curve is generated using the bootstrap resampling method (n = 1000 times), and the confidence interval is calculated using the Delong test.

10. An early screening system for ASD based on the measurement of refractive state and eye biological parameters according to claim 1, characterized in that, When it is determined as a high risk, the system automatically generates an intervention guide including a refractive correction plan (hyperopia compensation formula: ADD = 0.75×SE) and sensory integration training suggestions.

Images