Multi-domain eye-tracking for social attention assessment

Abstract

A social attention assessment system and method that provides objective developmental evaluation in children aged 12-48 months through multi-domain eye-tracking analysis. Examples of the system administer six brief tests measuring visual social attention, auditory social attention, and attention-shifting capabilities using naturalistic video stimuli and gaze-contingent audio presentation wherein infant-directed speech or non-social sounds are activated based on real-time gaze location. An eye-tracking device captures fixation patterns, saccadic movements, and attention-shifting cycles during stimulus presentation. Extracted metrics are compared against an age-gender stratified normative database containing data from over 2,000 children organized in 4-month age bins. Dual classification approaches including threshold-based and machine learning methods generate percentile rankings indicating where the child's social attention falls relative to age-matched peers. The complete assessment requires 10-20 minutes by trained technicians, enabling routine pediatric screening while providing detailed developmental profiles supporting intervention planning and preschool readiness assessment.

Description

Docket No. 1133.124WO1MULTI-DOMAIN EYE-TRACKING FOR SOCIAL ATTENTION ASSESSMENTCLAIM OF PRIORITY

[0001] This patent application claims the benefit of priority to U.S.Provisional Application Serial No. 63 / 755,197, filed February 6, 2025, which is incorporated by reference herein in its entirety.STATEMENT OF GOVERNMENT SUPPORT

[0002] This invention was made with government support under grant numbers R01-MH080134, R01-MH118879, and R37-MH132924 awarded by the National Institute of Mental Health. The government has certain rights in the invention,TECHNICAL FIELD

[0003] The subject matter disclosed herein relates to methods and systems for developmental assessment in children using eye-tracking technology, and more particularly to quantitative characterization of social attention abilities through automated gaze pattern analysis.BACKGROUND

[0004] Social attention is the ability to orient toward and process socially relevant stimuli such as faces, human speech, and social gestures. Social attention is fundamental to early childhood development. Developmental assessment in young children traditionally relies on clinical evaluation methods conducted by licensed psychologists, which typically requires several hours to complete. Parent-report screening tools offer rapid administration but demonstrate high false-positive rates, with pediatricians referring only approximately 35% of children who fail such screenings due to lack of confidence in results.Docket No. 1133.124WO1

[0005] To improve developmental assessment, eye-tracking technology has been explored for objective measurement of attention patterns in children. Existing eye-tracking techniques for developmental assessment include single-domain methods that assess only visual social attention through brief video presentations, as well as extended single-video approaches that analyze spatial gaze trajectories across 10-15 minute viewing periods.Recent advancements utilize consumer-grade hardware, including iPad-based systems that determine gaze direction by analyzing head position. These types of solutions, however, prioritize accessibility at the expense of accuracy.

[0006] There are several technical challenges facing current eye-tracking approaches. Single-domain assessments demonstrate limited sensitivity, with published studies reporting detection rates of 17-29% for children with developmental concerns. Single-metric classification approaches fail to leverage complementary information from multiple behavioral indicators. Creating meaningful normative comparisons requires large datasets with sufficient sample sizes to account for age-related developmental changes and individual variability, yet many existing approaches rely on normative samples of fewer than 200 participants.

[0007] While performing multiple eye-tracking tests usually improves assessment accuracy, combining multiple eye-tracking tests presents additional challenges. Uncertainty persists about which tests to combine, how to weight the results, and how to integrate diverse data types.Furthermore, eye-tracking systems generate complex numerical data that can be difficult for parents and clinicians to interpret and use in making decisions about a child's care.SUMMARY

[0008] Disclosed herein are systems and methods for assessing social attention in young children aged 12 to 48 months using eye-tracking technology and normative percentile-based profiling. Examples of the social attention assessment system and method administers a multi-domain testDocket No. 1133.124WO1battery comprising visual social attention tests, auditory social attention tests, and shifting social attention tests, each measuring distinct aspects of a child's spontaneous attention to social stimuli. The visual tests present paired social and non-social images, the auditory tests employ gaze-contingent technology wherein infant-directed speech or non-social sounds are activated based on real-time gaze location, and the shifting tests measure attention following in response to joint attention probes with pointing gestures. Eyetracking hardware captures gaze patterns during brief test presentations totaling approximately 10 to 20 minutes, generating quantitative metrics including fixation percentages, saccadic movements, and attention-shifting cycles.

[0009] The system and method compare extracted eye-tracking metrics to a large normative reference database containing data from over 2,000 children stratified by age bins of 4-month intervals and by gender. This stratified organization enables precise age-matched and gender-matched comparisons, accounting for rapid developmental changes during toddlerhood and systematic sex differences in social attention patterns. The system generates percentile rankings for each metric indicating where the child's performance falls relative to age-matched and gender-matched peers, providing continuous quantitative characterization rather than binary classifications. Dual classification approaches including both simple-threshold methods and machine learning models provide complementary analytical perspectives, with validation data demonstrating high specificity and positive predictive values for developmental assessment.

[0010] The multi-domain architecture captures heterogeneous social attention patterns that single-domain assessments fail to detect, as different children exhibit weaknesses in different domains. The complete assessment provides clinically interpretable reports with domain-specific percentile rankings, graphical visualizations, and identification of children with multiple-domain impairments who show exceptionally high likelihood of requiring developmental intervention. The rapid administration by trained technicians, combined with objective quantitative measurement and detailedDocket No. 1133.124WO1normative comparison, enables deployment across pediatric offices, preschools, and research settings for applications including preschool readiness assessment, intervention planning, and progress monitoring during the critical developmental window when early identification and intervention are most effective.BRIEF DESCRIPTION OF THE DRAWINGS

[0011] To easily identify the discussion of any particular element or act, the most significant digit or digits in a reference number refer to the figure number in which that element is first introduced.

[0012] FIG. 1 illustrates a comprehensive multi-domain percentile-based social attention assessment system for evaluating social attention across multiple domains in young children.

[0013] FIG. 2 illustrates the organization and composition of a multi-domain eye-tracking test battery for comprehensive social attention assessment in young children.

[0014] FIG. 3 illustrates a gaze-contingent audio delivery system for measuring auditory social attention preferences in young children through real-time control of audio stimulus presentation based on the child's gaze location,

[0015] FIG. 4 illustrates the architecture and organization of a normative reference database that facilitates percentile-based social attention profiling.

[0016] FIG. 5 illustrates a detailed flowchart of the overall method for assessing social attention in young children using eye-tracking technology.

[0017] FIG, 6 illustrates a diagrammatic representation of a computing system within which instructions may be executed for causing a machine to perform any one or more of the methodologies discussed herein.Docket No. 1133.124WO1DETAILED DESCRIPTION

[0018] The following description of examples of the invention is not intended to limit the invention to these examples, but rather to enable any person skilled in the art to make and use this invention.General Overview

[0019] The examples of the social attention assessment system and method described herein are an automated eye-tracking system that generates percentile-based reports characterizing a child's social attention abilities. The system and method administers six brief (58-95 second) eye-tracking tests to toddlers aged 12-48 months and produces percentile rankings showing where each child's performance falls relative to over 2,000 age-matched and gender-matched peers. Unlike diagnostic tools that provide binary classifications, this invention delivers a continuous developmental profile indicating relative strengths and weaknesses in social attention - skills critical for early learning, language development, and social interaction.

[0020] Traditional assessments by licensed psychologists require several hours, are expensive, and rely on subjective clinical judgment. Parent-report screening tools are faster but suffer from high false-positive rates and lack the granular detail needed to identify specific areas of concern. Social attention, which encompasses visual attention to faces and social scenes, auditory attention to human speech, and ability to shift attention in response to social cues, develops rapidly during the toddler years with tremendous individual variation.

[0021] Examples of the system and method use six complementary eyetracking tests assessing three distinct domains: (1) visual social attention through paired presentations of social versus non-social images; (2) auditory social attention through gaze-contingent paradigms controlling whether children hear infant-directed speech or non-social sounds; and (3) shifting social attention through joint attention tests measuring ability to follow pointing gestures. This multi-domain structure captures developmental heterogeneity, as children may show domain-specific strengths andDocket No. 1133.124WO1weaknesses. The auditory tests use a new gaze-contingent technology that monitors real-time gaze location and dynamically controls audio presentation — when the child fixates on one area, infant-directed speech plays; when gaze shifts elsewhere, non-social audio activates. The system and method also include a large-scale age-gender stratified normative reference database. This normative reference database contains eye-tracking data from over 2,000 neurotypically developing children aged 12-48 months, systematically stratified by 4-month age bins and gender. The system and method also includes an automated percentile-based reporting system that automatically generates reports transforming raw eye-tracking metrics into clinically interpretable percentile rankings.

[0022] Beyond simple fixation measurements, examples of the system and method quantify saccadic eye movement patterns reflecting exploration strategies, fixation durations indicating processing depth, latency to first social fixation, alternations between social and non-social stimuli, and triadic attention cycles (target-face-target sequences) demonstrating coordinated social attention. This multi-metric approach provides characterization beyond single-measure assessments. The complete test battery administers in approximately 12-15 minutes by trained technicians rather than doctoral-level clinicians. Tests employ naturalistic, engaging stimuli (brief child-friendly movies) maintaining toddler attention without requiring explicit instructions. This enables deployment across pediatric offices, preschools, and research laboratories. Moreover, the percentile-based characterization supports preschool readiness assessment, intervention planning by identifying specific domains requiring support, progress monitoring through repeated assessments tracking developmental trajectories, research examining relationships between social attention and other developmental variables, and early identification of children with social attention weaknesses (below 25th percentile) who may benefit from speech therapy or social skills training, regardless of specific diagnostic status.Docket No. 1133.124WO1

[0023] Examples of the social attention assessment system and method address fundamental limitations in developmental assessment by completing comprehensive evaluation in 10-20 minutes through trained technicians rather than requiring 2-3 hours with doctoral-level clinicians, dramatically increasing accessibility and enabling routine screening in primary care settings. The multi-domain test battery architecture solves the critical problem that single-domain approaches fail to capture heterogeneous social attention patterns. Validation data demonstrates sensitivity improves from approximately 20% (single test) to 55-69% (six-test battery) while maintaining high specificity (90-97%) across over 2,400 children. Different children fail different tests depending on their specific pattern of impairment. By assessing visual, auditory, and shifting attention domains, the examples of the system identifies children who would be missed by single-domain methods because their social attention challenges are domain-specific.

[0024] The gaze-contingent audio delivery system represents a technical achievement that measures active, sustained preference for social speech rather than passive listening responses. By enabling children to volitionally control audio presentation through gaze location, with audio switching occurring within 300-500 milliseconds, examples of the system captures genuine auditory social attention preferences that predict language development. This required specialized engineering to filter noisy toddler eye-tracking data, implement hysteresis preventing spurious switching, and coordinate real-time control across hardware and software components. The dual non-social audio controls (traffic sounds versus electronic sounds) provide methodological rigor through convergent evidence, with validation data from over 700 children per test demonstrating high classification accuracy (sensitivity 26-31%, specificity 97%, positive predictive value 94%).

[0025] The age-gender stratified normative database containing data from over 2,000 children enables percentile-based profiling that provides clinically richer characterization than binary classification approaches. TheDocket No. 1133.124WO1empirically optimized 4-month age bins capture meaningful developmental changes during toddlerhood while maintaining adequate statistical power within bins. Finer binning degrades performance through reduced sample sizes, while wider binning groups children at different developmental stages. Gender stratification addresses systematic sex differences in social attention development (effect sizes 0.2-0.4), improving precision of percentile ranking and avoiding inappropriate comparisons that would systematically misestimate male children's performance. The discovery that children failing three or more tests show 100% positive predictive value for autism represents an unexpected finding enabling rule-in diagnosis with exceptional confidence, allowing children with severe multi-domain impairments to be fast-tracked for treatment without awaiting lengthy comprehensive evaluations.System Architecture Overview

[0026] FIG, 1 illustrates a comprehensive multi-domain percentile-based social attention assessment system 100 for evaluating social attention across multiple domains in young children. The social attention assessment system 100 provides an objective, quantitative platform for characterizing social attention abilities in toddlers aged 12 to 48 months by measuring gaze patterns during presentation of social and non-social stimuli across visual, auditory, and attention-shifting domains. The social attention assessment system 100 generates percentile rankings that indicate where a child's social attention performance falls relative to age-matched and gender-matched normative data from a large reference population.

[0027] At the highest organizational level, FIG. 1 depicts the multi-domain percentile-based social attention assessment system 100 as an integrated assessment platform comprising four functional layers that work in cooperation. These four functional layers include hardware components 105 for stimulus presentation and gaze capture, a computing system 110 for data processing and analysis, data storage 115 for normative reference data and patient records, and an output / interface 120 for clinical reporting and dataDocket No. 1133.124WO1visualization. These layers are interconnected through defined data pathways that enable real-time gaze monitoring, stimulus control, normative comparison, and result generation.Hardware Components

[0028] The hardware components 105 include the physical devices necessary for presenting stimuli and capturing eye-tracking data. This layer includes three primary components that work in coordination: one or more of an eye-tracking device 125, one of more of a display device 130, and an audio output system 135.

[0029] The eye-tracking device 125 forms the gaze measurement component of the social attention assessment system 100. In some examples, the eye-tracking device 125 comprises a commercially available eye-tracking system, such as those manufactured by Tobii AB (Danderyd, Sweden). In alternative examples, other eye-tracking systems having similar capabilities can be employed. The eye-tracking device 125 includes a camera sensor for capturing images of the child's eyes, an infrared (IR) illuminator for generating corneal reflections that facilitate gaze estimation, and gaze estimation circuitry or software that calculates point-of-gaze coordinates on the display device 130 based on pupil position and corneal reflection positions. In some aspects of the disclosure, the eye-tracking device 125 operates at a sampling rate of at least 60 Hz, or more specifically 60 to 600 Hz, or even more specifically 120 Hz, to capture temporally precise gaze data during rapid eye movements characteristic of toddler viewing behavior. The eye-tracking device 125 is positioned at or integrated with the display device 130, typically below the display screen. In some examples, the eyetracking device 125 is positioned at a distance of approximately 60 to 70 cm from the child's eyes.

[0030] The display device 130 presents visual stimuli to the child during test administration. The display device 130 can include a computer monitor, LCD screen, or other suitable display having sufficient size and resolution to present stimuli clearly to toddlers at typical viewing distances. In oneDocket No. 1133.124WO1example, the display device 130 comprises a monitor having a screen size of at least 17 inches diagonal, or in other examples 20 to 24 inches, with resolution of at least 1024×768 pixels. The display device 130 is controlled by the computing system 110 to present test stimuli in a standardized sequence, wherein in one example visual stimuli are displayed for predetermined durations ranging from 58 to 95 seconds per test, depending on the specific test being administered.|0031 ] The audio output system 135 delivers auditory stimuli during specific tests that assess auditory social attention. The audio output system 135 comprises speakers positioned adjacent to or integrated with the display device 125, audio amplification circuitry, and volume control mechanisms. The speakers are positioned to provide clear audio delivery to the child without requiring the child to turn away from the display. In some implementations that utilize gaze-contingent audio presentation (as described in greater detail with reference to FIG. 3), the audio output system 135 is capable of rapidly switching between different audio streams with latency of less than 500 milliseconds, and in some cases less than 300 milliseconds, based on real-time gaze location monitoring.

[0032] The hardware components 105 are configured for use with toddlers in a clinical or research setting. During test administration, the child sits on a parent's lap or in an appropriate child seat positioned at the standardized distance from the display device 130, while the eye-tracking device 125 continuously monitors gaze position and the display device 130 and audio output system 135 present stimuli according to the test protocol.Computing System

[0033] The computing system 110 provides the data processing, analysis, and control functions that enable the multi-domain percentile -based social attention assessment method. The computing system 110 includes a personal computer, workstation, server, or other computing device having sufficient processing capability to perform real-time gaze tracking, stimulus presentation control, feature extraction, classification, and report generation.Docket No. 1133.124WO1in some examples, the computing system 110 is a standard desktop or laptop computer running a Windows, Mac OS, or Linux operating system.

[0034] The computing system 110 includes several components that work in cooperation, A processor 142 executes software instructions to perform computational tasks. The processor 140 can include a central processing unit (CPU), graphics processing unit (GPU), or combination thereof, having clock speeds typically ranging from 2 to 4 GHz or higher. Multiple processor cores may be employed to enable parallel processing of gaze data streams and stimulus presentation.

[0035] Memory 145 stores data and instructions during active processing. The memory 145 includes random access memory (RAM) or other volatile memory having sufficient capacity to hold software instructions, incoming gaze data streams, normative database query results, and intermediate processing results. In one example, the memory 145 includes at least 8 GB of RAM, or more in alternative examples, such as 16 to 32 GB or more.

[0036] Storage 150 provides persistent data storage for software applications, normative databases, patient data, and system configuration information. The storage 150 may comprise hard disk drives, solid-state drives, or other non-volatile storage media having capacity of at least 500 GB or more to accommodate the normative database and accumulated patient records.

[0037] A network interface 155 enables communication with external systems and remote access capabilities. The network interface 155 can be a wired Ethernet connectivity, wireless networking (WiFi), or both, enabling the social attention assessment system 100 to access cloud-based databases, transmit results to electronic health record systems, or enable remote test administration and monitoring.

[0038] Stored within memory 145 and executed by the processor 140 are software instructions 160 that implement the assessment functionality. TheDocket No. 1133.124WO1software instructions 160 are organized into several functional components that work in cooperation to perform the assessment method. It should be understood that while these components are described as distinct functional elements, they may be implemented as integrated software code, separate software modules, libraries, or services, and may be written in any suitable programming language such as Python, C++, Java, or combinations thereof.

[0039] A test battery 165 manages the presentation sequence of eye-tracking tests. The test battery 165 coordinates the display of visual stimuli on display device 130 and delivery of audio stimuli through audio output system 135 according to standardized test protocols. Tire test battery 165 presents tests in a predetermined or randomized sequence, implements timing controls to ensure each test runs for its specified duration, and implements gaze-contingent stimulus control for auditory tests wherein audio selection is determined by real-time gaze location. The test battery165, in cooperation with the data processing component 170, implements gaze-contingent audio delivery during auditory social attention tests. The auditory social attention tests are described below in connection with FIG, 2 and the gaze-contingent audio delivery is described in greater detail below with reference to FIG. 3. Real-time gaze location monitoring controls audio stimulus selection with, in some examples, a switching latency of less than 500 milliseconds. The test battery 165 also implements quality control functions including monitoring total looking time during each test, detecting calibration drift, and flagging tests that fail to meet minimum data quality thresholds.

[0040] A data processing component 170 extracts quantitative eye-tracking metrics from the raw gaze data streams received from eye-tracking device 125. The data processing component 170 performs several operations including filtering raw gaze coordinates to remove blinks and artifacts, mapping gaze coordinates to areas of interest (AOIs) defined for each test stimulus, calculating fixation events and durations, computing saccade (rapid eye movement) frequencies, determining percent fixation time within each AOI, and calculating attention-shifting patterns such as triadic attentionDocket No. 1133.124WO1cycles in the joint attention test. For each administered test, the data processing component 170 extracts between 6 and 9 distinct eye-tracking metrics that quantify different aspects of social attention performance. These metrics include, but are not limited to: (i) percent fixation on social stimuli; (ii) percent fixation on non-social stimuli; (iii) number of saccades per second within social AOIs; (iv) number of saccades per second within nonsocial AOIs; (v) percent fixation on eyes; (vi) percent fixation on face; (vii) and number of joint attention cycles.

[0041] A classification engine 175 applies analytical methods to the extracted metrics to generate social attention assessments. The classification engine 175 implements dual classification approaches: a simple -threshold classifier and a machine learning classifier. The simple-threshold classifier compares extracted metrics to predetermined threshold values that have been empirically derived to achieve high specificity (at least 97%) in distinguishing typical from atypical social attention patterns. For visual and auditory social attention tests, the threshold is set at 69% fixation on non¬ social stimuli. For the joint attention test, the threshold is 30% fixation on non-target objects. One the other hand, the machine learning classifier comprises a trained statistical model, such as a logistic regression model, random forest model, or other supervised learning algorithm, which has been trained on a large dataset of children with known diagnostic outcomes to predict social attention classification based on multiple input features. In some examples, the classification engine 175 generates classification scores ranging from 0 to 100, wherein higher scores indicate greater deviation from typical social attention patterns. In other examples, higher scores can indicate elevated likelihood of autism spectrum disorder or other developmental conditions affecting social attention.

[0042] A report generator 178 produces assessment reports and visualizations communicating results to clinicians and caregivers. The report generator 178 compiles results from each administered test, retrieves age- matched and gender-matched normative reference values from a normativeDocket No. 1133.124WO1database 180, calculates percentile rankings for each metric by comparing the child’s performance to the reference distribution, generates textual summaries of domain-specific performance (visual social attention, auditory social attention, shifting social attention), creates graphical visualizations showing where the child falls on normative distributions, and formats the compiled information into a clinical report suitable for interpretation by healthcare providers and parents. The report generator 178 can produce reports in various formats including PDF documents, HTML pages, or formatted data streams for integration with electronic health record systems.Data Storage

[0043] A data storage 185 provides persistent storage for normative reference data and patient assessment data. In some examples, the data storage 185 includes two logically distinct databases that may be physically stored on storage 150, on network-attached storage, or in cloud-based database systems.

[0044] The normative database 180 stores eye-tracking data from a large reference population of neurotypically developing children. In one example, the normative database 180 contains data from at least 2,000 children, and in some examples at least 2,200 children, and in other examples between 2,400 and 3,000 children, aged 12 to 48 months who completed one or more of the six eye-tracking tests during well-child visits or research participation. The normative database 180 is organized with age stratification 186 wherein children are grouped into age bins, typically of 4-month interval (e.g., 12- 15 months, 16-19 months, 20-23 months, etc.), and with gender stratification 188 wherein separate normative values are maintained for male and female children within each age bin. For each metric extracted from each test, the normative database 160 stores statistical summary- values including mean, standard deviation, and percentile distributions (e.g., 5th, 10th, 25th, 50th, 75th, 90th, 95th percentiles). This stratified organization enables precise age-matched and gender-matched comparisons when assessing an individual child, accounting for developmental changes that occur rapidly during the toddler period.Docket No. 1133.124WO1

[0045] The patient data store 190 houses assessment data for individual children who have undergone testing. For each assessed child, the patient data store 190 stores raw gaze data captured during test administration, extracted eye-tracking metrics calculated by the data processing component 170, classification results from the classification engine 175, and generated assessment reports. The patient data store 190 can also store demographic information, test session metadata (date, time, testing location, technician), and quality indicators for each test session. The patient data store 190 enables longitudinal tracking of individual children who undergo repeated assessments, facilitating evaluation of test-retest reliability and monitoring of developmental trajectories over time.Output / interface

[0046] The output / interface 192 provides mechanisms for communicating assessment results to end users and integrating with external information systems. An assessment report 194 comprises a formatted document presenting the child's performance across tested domains. The assessment report 194 includes domain-specific scores showing percentile rankings for visual social attention, auditory social attention, and shifting social attention. The assessment report 194 also can include individual test results showing raw metric values and percentile rankings for each completed test; classification results from both the simple-threshold and machine learning approaches, as well as graphical visualizations depicting where the child falls on normative distribution curves. The assessment report 194 further can show clinical interpretation text that may flag performance falling below the 25th percentile as potentially indicating need for further evaluation or intervention. The output / interface 192 generates the assessment report 194 in a format suitable for inclusion in medical records and for review bypediatricians, psychologists, speech therapists, or other professionals involved in the child's care.

[0047] A clinical dashboard 196 provides an interactive visualization interface for reviewing the assessment results 194. The clinical dashboardDocket No. 1133.124WO1196 can be implemented as a graphical user interface displayed on a computer screen, tablet, or other display device, enabling users to view detailed test results, compare performance across multiple assessment sessions, examine raw gaze data visualizations, and access population-level statistics from the normative database 180. The clinical dashboard 196 facilitates clinical research applications by enabling researchers to query’ patient data, generate aggregate statistics, and export data for statistical analysis.

[0048] External systems integration 198 enables the social attention assessment system 100 to communicate with other information systems. In some examples, external systems integration 198 includes connectivity to electronic health record (EHR) systems, enabling automatic transmission of assessment reports to a child’s medical record maintained by their healthcare provider. In other examples, the external systems integration 198 includes connectivity to research databases, enabling de-identified assessment data to be contributed to multi-site research studies or registries. The external systems integration 198 can utilize standard healthcare data exchange protocols such as HL 7 or FHIR to ensure interoperability.Alternative Embodiments and Optional Features

[0049] Several alternative implementations and optional features may be implemented within the system architecture of FIG. 1. In one alternative example, the computing system 110 includes a distributed system wherein some processing occurs on a local device (such as a computer at the testing site) while other processing occurs on a remote server accessed via network interface 155. For example, stimulus presentation and raw gaze data collection may occur locally, while feature extraction, classification, and report generation occur on a cloud-based server having access to a centrally maintained normative database 180.

[0050] In another alternative examples, the display device 130 includes a tablet computer having an integrated camera that approximates eye-tracking capability, enabling home-based or mobile deployment of the assessment, itDocket No. 1133.124WO1should be noted, however, that such configurations typically achieve lower gaze estimation accuracy compared to dedicated eye-tracking hardware. One optional feature that may be incorporated include pupillometry measurements wherein the eye-tracking device 125 measures pupil diameter changes as an index of autonomic arousal. Another optional feature can be the integration of parent questionnaire data that supplements the eyetracking metrics with parent- reported developmental concerns. Yet another optional feature is taking a video recording of the child's face and behavior during testing for later review. Another optional feature is the real-time data quality visualization enabling the technician to monitor calibration stability and data quality during test administration.

[0051] The software instructions 160 may optionally implement adaptive test selection wherein, based on performance on initial tests, examples of the social attention assessment system 100 select subsequent tests most likely to provide diagnostic information for that particular child. This can potentially reduce total testing time while maintaining classification accuracy. The normative database 180 may be implemented as a dynamic, continuously updated database wherein new validated data from assessed children is periodically incorporated, the age bin and gender stratifications are maintained, and statistical reference values are recalculated to improve precision over time. This living database approach enables the system to accumulate larger reference samples and potentially to adapt to population-level changes in social attention patterns.Multi-Domain Eye-Tracking Test Battery

[0052] FIG. 2 illustrates the organization and composition of a multi¬ domain eye-tracking test battery 200 for comprehensive social attention assessment in young children. The multi-domain eye-tracking test battery 200 illustrates specific examples of the test battery' 165 described in FIG.l, with FIG. 2 setting forth specific tests and test battery characteristics. The multi-domain eye-tracking test battery 200 provides a structured framework for measuring social attention capabilities across three distinct cognitive domains: visual social attention, auditory social attention, and shifting socialDocket No. 1133.124WO1attention. This is accomplished in part through administration of one or more of six complementary tests that collectively characterize a child's preference for social stimuli, responsiveness to social communication, and ability to follow social cues.

[0053] At the organizational level, the multi-domain eye -tracking test battery 200 includes a collection of six standardized eye-tracking tests systematically organized into three domains based on the modality and cognitive process being assessed. This multi-domain architecture addresses a fundamental challenge in developmental assessment, namely, that autism spectrum disorder and other conditions affecting social attention are heterogeneous. This means that different children exhibit different patterns of social attention strengths and weaknesses. By assessing multiple domains, the multi-domain eye-tracking test battery 200 captures this heterogeneity and provides comprehensive characterization that would not be achievable through assessment of a single domain. The three -domain structure organization also aligns with established diagnostic criteria for autism spectrum disorder, which specify deficits in social attention to faces and people (visual domain), reduced responsiveness to speech and social vocalizations (auditory’ domain), and impaired joint attention and social referencing (shifting domain), as defined in the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition (DSM-5).Visual Social Attention Domain

[0054] The multi-domain eye-tracking test battery’ 200 includes a visual social attention domain 210 that assesses a child's preferential allocation of attention to social visual stimuli as compared to non-social visual stimuli. This visual social attention domain 210 includes three tests that employ a preferential-looking paradigm, which is also known as a forced-choice paradigm. This paradigm presents social and non-social stimuli simultaneously in a side-by-side arrangement. The social attention assessment system 100 measures the child's spontaneous gaze preferences to determine relative attention allocation. The three tests within this visualDocket No. 1133.124WO1social attention domain 210 vary systematically in the complexity and dynamism of the social content, ranging from simple human movement to complex social interactions, thereby capturing social attention preferences across a range of social contexts.

[0055] The three tests of the visual social attention domain 210 include a Geometric Preference (abbreviated “GeoPref”) test 215, a Complex Social test 220, and an Outside Play test 225. In some examples, the GeoPref test 21 presents 28 non-repeating paired presentations of dynamic social images and dynamic geometric images displayed side-by-side in a linked fashion. The term "linked fashion" indicates that both stimuli are presented simultaneously and continuously throughout the test duration, enabling the child to freely allocate attention between the two stimulus categories. The social images depict human activity, such as children or adults engaged in movements like yoga poses, exercise, or simple actions. The geometric images comprise dynamic geometric patterns, such as rotating shapes, expanding / contracting patterns, kaleidoscope effects, or screensaver-style animations that provide visually engaging non-social control stimuli matched in motion and visual complexity to the social stimuli.

[0056] In some implementations, the GeoPref test 215 has a duration of between 58 and 66 seconds. This duration is selected to provide sufficient observation time to obtain stable estimates of attention allocation while remaining within the attention span limitations of toddlers aged 12 to 48 months. The 28 paired presentations are each shown for approximately 2 to 3 seconds before transitioning to the next pair, preventing habituation while accumulating sufficient data across multiple exemplars of social and geometric stimuli.

[0057] The social attention assessment system 100 presents the GeoPref test 215 with silent visual stimuli, meaning no audio accompanies the visual presentations. This isolation of the visual modality enables pureDocket No. 1133.124WO1measurement of visual social attention without confounding effects of auditory information that might direct attention.

[0058] During administration of the GeoPref test 215, the eye-tracking device 125 extracts between 6 and 12 metrics, or in some example 9 metrics, which quantify aspects of visual social attention. These metrics include: (i) percent fixation on social images, calculated as the total duration of fixations falling within social image areas of interest divided by total looking time for the test; (ii) percent fixation on non-social (geometric) images, calculated analogously; (iii) number of saccades per second within social image AOIs, indicating frequency of exploratory eye movements when viewing social content; (iv) number of saccades per second w ithin non-social image AOIs; (v) mean fixation duration on social images; (vi) mean fixation duration on non-social images; (vii) latency to first fixation on social images following each stimulus transition; (viii) number of alternations between social and non-social images; and (ix) total looking time as a percentage of total test duration, which serves as a data quality indicator. Alternative or additional metrics may include median fixation durations, fixation count, or other statistical summaries of gaze behavior,

[0059] The GeoPref test 215 has been extensively validated in a large sample exceeding 2,200 children and demonstrates high specificity (at least 98%) when using a threshold of 69% fixation on geometric images to classify atypical visual social attention. The GeoPref test 215 is particularly effective at identifying a subtype of autism characterized by strong preference for geometric patterns over social content.Complex Social Test

[0060] The Complex Social test 220 presents 9 non-repeating paired presentations of geometric patterns and complex social scenes. The Complex Social test 220 extends the GeoPref paradigm by presenting social stimuli depicting more sophisticated social interactions, such as disagreements between children, conflicts, cooperative play scenarios, or emotionallyDocket No. 1133.124WO1nuanced exchanges. These complex social scenes require more advanced social cognitive processing compared to the simple human movement depicted in the GeoPref test 215. This potentially engages neural systems involved in understanding social intentions, emotions, and interpersonal dynamics.|00611 In some examples, the Complex Social test 220 has a duration of between 90 and 100 seconds. This longer duration, as compared to the GeoPref test 215, reflects the more complex social content requiring longer viewing time to fully appreciate the social interactions. In some implementations, the 9 paired presentations are each displayed for approximately 10 seconds, allowing sufficient time for the child to process the more cognitively demanding social content. Similar to the GeoPref test 215, the Complex Social test 220 employs silent visual presentation to isolate visual attention. The geometric control stimuli in this test are matched to the social stimuli in visual complexity, motion, and interest level to provide an appropriate non-social comparison condition.

[0062] The Complex Social test 220 generates 9 eye-tracking metrics parallel to those extracted from the GeoPref test 215, enabling comparison of attention allocation patterns across tests of differing social complexity. In some examples, children who show typical social attention on the simpler GeoPref test 215 but reduced social attention on the Complex Social test 220 may have specific difficulties with processing higher-order social interactions, providing diagnostically relevant information about the nature of social attention challenges.Outside Play Test

[0063] The Outside Play test 225, in some configurations, presents 10 nonrepeating paired presentations of dynamic fractal patterns and social scenes depicting high-action activities in large peer groups. The social scenes show children engaged in vigorous group activities such as playing tag, ring- around-the-rosie, running together, or other dynamic peer play scenariosDocket No. 1133.124WO1involving multiple children in motion. These stimuli assess attention to peer social contexts and group dynamics, which are particularly relevant for preschool readiness and social integration capabilities.

[0064] In some examples, the Outside Play test 225 has a duration of between 68 and 76 seconds. Each of the 10 paired presentations display for approximately 7 seconds, balancing the need for sufficient viewing time with the goal of presenting multiple exemplars of outdoor peer play contexts. The non-social control stimuli comprise dynamic fractal patterns, also known as geometric fractals or mathematical fractals, which are visually complex, self-similar patterns that provide engaging visual motion without social content. Fractals are specifically selected as controls for the high-motion social scenes because they provide matched visual complexity and motion without any biological or social features.

[0065] The Outside Play test 225 employs silent presentation and generates 9 eye-tracking metrics analogous to the other visual domain tests, enabling within-domain comparison of social attention across contexts ranging from individual humans (such as the GeoPref test 215), to dyadic / small group interactions (such as the Complex Social test 220), to large group peer activities (such as the Outside Play test 225).Characteristics of the Visual Social Attention Domain

[0066] The three tests of the visual social attention domain 210 collectively provide comprehensive assessment of visual social attention through a progression of social complexity. All three tests employ the preferential looking paradigm with side-by-side presentation, all are presented without audio to isolate visual attention, all extract the same set of 9 standardized metrics to enable cross-test comparison, and all use the same classification threshold of 69% fixation on non-social stimuli as indicative of reduced visual social attention. However, the tests differ in social content complexity (simple movement → complex interactions → large group dynamics), enabling identification of children who show domain-specific patterns suchDocket No. 1133.124WO1as preserved attention to simple social content but reduced attention to complex social situations.Auditory Social Attention Domain

[0067] The multi-domain eye-tracking test battery 200 also includes an auditory social attention domain 230 that assesses a child's preferential allocation of attention to social auditory stimuli, specifically human speech and vocalizations, as compared to non-social auditory stimuli. The auditory social attention domain 230 employs a unique gaze-contingent paradigm wherein the child's real-time gaze location determines which audio stimulus is delivered, enabling measurement of auditory social attention preferences through the child's volitional control of their auditory environment. The two tests within this auditory social attention domain 230 provide convergent measurement of speech preference using different non-social audio controls, enhancing reliability and generalizability of auditory’ social attention assessment. The two tests of the auditory social attention domain 230 include a Motherese vs. Highway test 235 and a Motherese vs. Techno test 240.Motherese vs. Highway Test

[0068] The Motherese vs. Highway test 235 presents a gaze-contingent paradigm contrasting infant-directed speech against traffic noise. The term "motherese," which is also known as infant-directed speech, child-directed speech, or “parentese,” refers to the specialized speech register that caregivers naturally use when addressing infants and toddlers. This type of speech is characterized by elevated pitch, exaggerated prosody, slow tempo, repetitive phrases, high question intonation frequency, and emotional expressiveness. In the Motherese vs. Highway test 235, an actress is recorded telling a story to a teddy bear using motherese vocal characteristics, creating a social audio stimulus that is ecologically valid and naturally engaging to typically developing toddlers.Docket No. 1133.124WO1|00691 The non-social audio stimulus comprises traffic sounds, also referred to as highway sounds or road noise, including recordings of car engines, horn honks, vehicles passing, and ambient road noise. Traffic sounds are selected as the non-social control because they provide complex auditory stimulation with temporal variation and amplitude modulation (similar to speech) but lack any human vocal or communicative content.

[0070] The Motherese vs. Highway test 235 uses gaze-contingent technology wherein two visual target areas are displayed on the screen, and the child's fixation location determines audio delivery. When the child's gaze fixates on a first area of interest (such as the left side of the screen), the motherese audio stream is activated and plays through the speakers. When the child's gaze shifts to a second area of interest (such as the right side of the screen), the motherese audio is muted and the traffic sound audio stream is activated. In some configurations the audio switching occurs with latency of less than 200 milliseconds, from detection of gaze location change. This allows responsive auditory feedback that allows the child to exert volitional control over their auditory environment.

[0071] In some implementations, the Motherese vs. Highway test 235 has a duration of between 58 and 65 seconds. Throughout this duration, the child can freely shift their gaze between the two target areas, and the social attention assessment system 100 continuously monitors gaze location and controls audio delivery accordingly. Typically developing children show strong preference for motherese speech, spending 70% to 95% of looking time fixating on the area that triggers motherese audio. Children with reduced auditory social attention, conversely, may show weaker preference or even preference for the traffic sounds.

[0072] Prior to administering the Motherese vs. Highway test 235 (and the Motherese vs. Techno test 240 described below), the social attention assessment system 100 implements a gaze-contingent training procedure, also known as the Dog-Cat Training paradigm, to familiarize the child withDocket No. 1133.124WO1the concept of gaze-contingent control. During this training, images of a puppy and kitten are displayed side-by-side, and fixating on the dog image triggers a bark sound while fixating on the cat image triggers a purr sound. Training continues until the child demonstrates at least 2, 3, or 4, alternating fixations between the images, indicating comprehension of the gaze¬ contingent mechanism. If the child does not spontaneously alternate within approximately 20 seconds, a technician administering the test may provide a verbal prompt such as " Look at the dog!" or " Look at the cat!" to encourage exploration.

[0073] In some examples, the Motherese vs. Highway test 235 generates 9 metrics including: (i) percent fixation triggering motherese audio; (ii) percent fixation triggering traffic audio; (iii) number of audio switches (gaze alternations between AOIs); (iv) mean duration of motherese listening bouts; (v) mean duration of traffic listening bouts; (vi) latency to first motherese fixation; (vii) total looking time; (viii) number of saccades during motherese periods; and (ix) number of saccades during traffic periods. These metrics collectively characterize the strength of preference for social speech and the child's engagement with auditory’ social stimuli,Motherese vs. Techno Test

[0074] The Motherese vs. Techno test 240 presents a gaze-contingent paradigm structurally identical to the Motherese vs. Highway test 235, but employs different non-social audio control stimuli. The social audio again comprises an actress telling a story using motherese vocal characteristics, providing consistency in the social stimulus across the two auditory tests and enabling assessment of test-retest reliability.

[0075] The non-social audio stimulus comprises electronic or technological sounds, also referred to as "techno sounds," synthesized tones, or abstract musical patterns. These sounds include synthesized musical notes, electronic beeps, digital sound effects, or computer-generated audio patterns that are acoustically interesting but lack any human vocal or communicative content.Docket No. 1133.124WO1Techno sounds are selected as an alternative non-social control to provide convergent evidence for auditory social attention preferences. If a child shows motherese preference over both traffic sounds (test 235) and techno sounds (test 240), then this convergent evidence strengthens confidence in the assessment of auditory social attention capability.|0076| In some configurations, the Motherese vs. Techno test 240 has a duration between 57 and 62 seconds. The Motherese vs. Techno test 240 employs the same gaze-contingent mechanism as the Motherese vs. Highway test 235, with audio switching latency of less than 500 milliseconds based on real-time gaze monitoring. The child's fixation on one target area activates motherese audio, while fixation on the other target area activates techno sounds, and the child can volitionally control audio selection through gaze allocation.

[0077] In some examples, the Motherese vs. Techno test 240 generates 9 metrics parallel to those of the Motherese vs. Highway test 235, substituting techno sound parameters for traffic sound parameters. The availability of two auditory social attention tests enables assessment of test-retest reliability within the auditory domain and provides redundancy such that if one auditory test yields poor data quality (such as due to child fussiness or technical issues), the other auditory test may still provide valid auditory social attention measurement.Auditory Domain Characteristics

[0078] The two tests of the auditory social attention domain 230 collectively provide robust assessment of auditory social attention through convergent measurement using different non-social controls. Both the Motherese vs. Highway test 235 and the Motherese vs. Techno test 240 employ gaze-contingent technology. This is a unique approach that is not present in existing developmental assessment methods, wherein the child actively controls stimulus presentation through gaze allocation rather than passively viewing predetermined stimulus sequences. This gaze-contingentDocket No. 1133.124WO1approach provides several advantages. First, engages the child's attention more effectively than passive viewing, second, it measures active preference rather than merely orienting responses, third, it enables quantification of how much time the child chooses to spend listening to social speech when given free choice, and fourth, it mimics naturalistic scenarios wherein children direct their attention to preferred auditory sources in their environment.

[0079] Both the Motherese vs. Highway test 235 and the Motherese vs. Techno test 240 employ the 69% threshold (percent fixation on non-social audio) as the criterion for identifying reduced auditory social attention, consistent with the visual domain tests. In some examples, both tests generate the same standardized set of 9 metrics, enabling within-domain comparison and aggregate auditory domain scoring. These two auditory tests require functional gaze-contingent implementation with low switching latency (typically 200-500 milliseconds) to provide responsive feedback to the child's gaze shifts, ensuring that the child experiences contingent control over the auditory environment.Shifting Social Attention Domain

[0080] The multi-domain eye-tracking test battery 200 also includes a shifting social attention domain 245 that assesses a child's ability to disengage from current attentional focus, follow social directives, and shift attention in response to another person's gestures and vocalizations. This shifting social attention domain 245 addresses joint attention capabilities, a critical developmental milestone that emerges in typically developing children between 9 and 18 months and is often impaired or delayed in children with autism spectrum disorder and other developmental conditions. The shifting social attention domain 245 includes a Joint Attention test 250.Joint Attention Test

[0081] The Joint Attention test 250 presents a video of an actress telling a story about a teddy bear while systematically initiating joint attention bids toDocket No. 1133.124WO1direct the child's attention to various objects in the scene. A joint attention bid, also known as an attention-directing gesture or social referencing cue, comprises a coordinated pointing gesture toward a target object accompanied by a verbal directive, typically the word "look," and often preceded or followed by eye gaze directed toward the target. During the test, the actress points to different objects positioned around the scene — such as a toy truck, a bow, the teddy bear itself, or other items — while saying "look" or similar attention-directing phrases.

[0082] In some examples, the Joint Attention test 250 initiates between 6 and 10 joint attention bids distributed across the test duration. Each joint attention bid is structured to include a clear pointing gesture, verbal directive, and sufficient pause to allow the child time to shift attention to the indicated target. The targets are positioned at various locations around the actress, requiring the child to shift gaze from the actress's face to peripheral locations, demonstrating the ability to follow social directives. In some configurations, the Joint Attention test 250 has a duration of between 85 and 100 seconds. This duration accommodates the 8 joint attention bids with sufficient spacing to allow measurement of responses while maintaining toddler engagement.

[0083] Unlike the visual and auditory tests which measure preferential allocation between two stimulus types, the Joint Attention test 250 measures attention-shifting behavior in response to specific social cues. The Joint Attention test 250 defines several area categories: target objects (items the actress points to), non-target objects (items present in the scene but not indicated), the actress's face, and background regions. The social attention assessment system 100 tracks where the child looks during and following each joint attention bid to determine whether the child successfully follows the social directive.

[0084] In some examples, the Joint Attention test 250 generates 9 metrics specifically adapted to measure attention-shifting capability. These 9 metricsDocket No. 1133.124WO1include: (i) percent fixation on target objects (objects indicated by the actress); (ii) percent fixation on non-target objects (distractors or items not indicated); (iii) percent fixation on the actress's face; (iv) number of complete triadic attention cycles, wherein a triadic cycle comprises a sequence of fixations moving from target to face to target again (Target-Face-Target) or from face to target to face again (Face -Target-Face), indicating coordinated attention between the social partner and the object of joint reference; (v) number of partial joint attention cycles (e.g., Face-Target without return); (vi) response latency, defined as time from initiation of joint attention bid to first fixation on target; (vii) total number of attention shifts during bid periods; (viii) mean fixation duration on targets; and (ix) total looking time.|0085| The Joint Attention test 250 employs a different classification threshold compared to the visual and auditory tests. Specifically, 30% or greater fixation on non-target objects (as opposed to 69% for other tests) is used as the threshold for identifying atypical shifting attention. This lower threshold reflects the different task structure. In preferential looking tests, typical children spend 70-90% time on social stimuli, leaving 10-30% on non-social stimuli. While in the joint attention test, typical children should be following the actress's points and spending most time on targets and face, with less than 30% time on non-indicated objects. The different threshold value is empirically derived to achieve similar high specificity (at least 95%, or more specifically at least 96%) as the visual and auditory tests.Test Battery Characteristics and Technical Specifications

[0086] The multi-domain eye-tracking test battery 200 also includes test battery characteristics 255 that summarize the overall parameters and specifications of the complete test battery. These parameters and specifications include one or more of the following: (i) total number of tests used in the test battery; (ii) total duration of the complete test battery; (iii) number of metrics per test; and (iv) duration of each individual test in the test batter}'-.Docket No. 1133.124WO1

[0087] In some examples, the multi-domain eye-tracking test battery 200 includes a total of 6 tests, specifically organized as 3 visual social attention tests, 2 auditory social attention tests, and 1 shifting social attention test, for a total of 6 tests. This 6-test composition provides comprehensive multi¬ domain coverage while remaining administratively feasible within typical clinical time constraints.

[0088] In some examples, the total duration for administering the complete text battery ranges from 10 to 20 minutes. This typically includes setup time, calibration, and test presentation. In some configurations, the durations of each individual test range from between 58 to 95 seconds. In some examples there is a mean duration of approximately 73 seconds per test. Actual session duration varies based on factors including child cooperation, calibration efficiency, and whether tests need to be repeated due to data quality concerns.

[0089] In some implementations, each test generates between 6 and 12 metrics, yielding between 12 and 72 metrics depending on the number of tests completed. In some examples, each test generates 9 metrics per test, yielding a total of 54 metrics when all six tests are completed. This multimetric approach provides rich characterization of social attention across multiple dimensions beyond simple preference measurements.

[0090] At least two tests in the multi-domain eye-tracking test battery 200 use gaze-contingent technology, specifically the Motherese vs. Highway test 235 and the Motherese vs. Techno test 240. This gaze-contingent capability, wherein stimulus presentation is controlled in real-time by the child's gaze location, represents a useful technical feature and enables active measurement of auditory social attention preferences. The multi-domain eye¬ tracking test battery 200 is designed as a multi-domain assessment that integrates measurements across visual, auditory, and shifting attention capabilities. This multi-domain approach addresses the heterogeneity ofDocket No. 1133.124WO1social attention profiles across children: some children may show reduced social attention primarily in the visual domain, others primarily in the auditory domain, and still others across multiple domains. By assessing all three domains, the battery captures this heterogeneity and maximizes sensitivity for detecting social attention weaknesses regardless of their specific pattern.

[0091] The test included in the multi-domain eye-tracking test battery 200 have been validated in a large, diverse sample exceeding 2,000 children representing multiple diagnostic categories including autism spectrum disorder, various non-autism developmental delays (language delay, global developmental delay, motor delay), and typical development. This validation sample provides the normative reference data stored in normative database 180 and establishes the empirical thresholds, expected metric distributions, and classification accuracy parameters for each test.Alternative Embodiments and Optional Features

[0092] Several alternative implementations variations of the multi-domain eye-tracking test battery 200 may be implemented while maintaining the core multi-domain assessment structure. In one alternative example, the multi-domain eye-tracking test battery 200 includes fewer than six tests, such as 2, 3, 4, or 5 tests selected from the six available tests, while still maintaining representation from at least two different domains. For example, a reduced battery might include one visual test, one auditory test, and the joint attention test, providing three-domain coverage in approximately 7-10 minutes total test duration. Sensitivity for detecting social attention weaknesses is reduced with fewer tests, but specificity typically remains high, making abbreviated batteries suitable for screening applications where time is limited.

[0093] In another alternative implementations, the multi-domain eyetracking test battery 200 includes additional tests beyond the six described, such as additional visual tests employing different social contexts (e.g.,Docket No. 1133.124WO1infant faces, animal images with social relevance), additional auditory tests employing different non-social controls (e.g., musical instruments, nature sounds), or additional attention-shifting tests employing different social cue modalities (e.g., gaze direction without pointing, reaching gestures). These additional tests may be developed and validated following the same paradigm structure and metric extraction approaches. Moreover, these additional test can be incorporated into the multi-domain framework to further improve sensitivity or to assess additional subdomains of social attention.

[0094] The specific stimuli used in each test may vary while maintaining the essential structure of social vs. non-social contrasts. For example, the GeoPref test 215 can alternatively use social stimuli depicting animals with social relevance (such as pets, familiar animals) or inanimate objects with faces drawn on them, as opposed to human actors. The Complex Social test 220 can employ different specific social interaction scenarios while maintaining the complexity level. The Outside Play test 225 can depict indoor group activities (such as circle time at preschool) rather than outdoor play. The motherese audio might be delivered by male actors in addition to or instead of female actors. Such stimulus variations maintain the core assessment paradigm while enabling cultural adaptation, avoiding habituation in repeated testing, or testing hypotheses about stimulus-specific effects,

[0095] The order of test presentation may be fixed, randomized, or adaptive. In a fixed sequence embodiment, tests are always presented in the same order (such as, for example, visual tests first, then auditory tests, then shifting tests). In a randomized sequence implementation, test order is randomized across children to control for order effects such as fatigue or learning. In an adaptive sequence configuration, later test selection is determined based on performance on earlier tests, such as selecting tests most likely to provide diagnostic information for children showing ambiguous results on initial tests.Docket No. 1133.124WO1

[0096] The metrics extracted from each test may be expanded or modified while maintaining the core set of measurements. For example, additional metrics might include pupillometry measurements (pupil diameter changes as an index of arousal or cognitive load), blink rate or blink pattern analysis, depth of processing indices based on fixation duration distributions, or scan path complexity measures. Alternative metrics might substitute for or supplement those described, such as using median rather than mean fixation durations, or calculating entropy measures of attention distribution.

[0097] It should be understood that features from different tests within the multi-domain eye-tracking test battery’ 200 can be combined or intermixed in various ways. For example, the gaze-contingent mechanism employed in the auditory tests could alternatively be applied to visual tests, such as presenting social images when the child looks at one location and geometric images when looking at another location. Similarly, the joint attention paradigm structure could be extended to include additional bids or to test following of gaze direction alone (without pointing), or pointing alone (without verbal directive), to isolate different components of joint attention capability. The specific threshold values (such as, by way of example, 69% for visual / auditory tests, 30% for joint attention) could be adjusted to prioritize different trade-offs between sensitivity and specificity depending on the intended application (e.g., screening vs. diagnosis). The structure of the multi-domain eye-tracking test battery 200 is sufficiently flexible to accommodate additional tests, alternative metrics, or modified paradigms while maintaining the fundamental principle of comprehensive three-domain social attention assessment.Gaze-Contingent Audio Stimulus Delivery System

[0098] FIG. 3 illustrates a gaze-contingent audio delivery system 300 for measuring auditory social attention preferences in young children through real-time control of audio stimulus presentation based on the child's gaze location. The gaze-contingent audio delivery’ system 300 facilitates objectiveDocket No. 1133.124WO1quantification of a child's preference for social auditory stimuli (such as infant-directed speech) versus non-social auditory stimuli (such as traffic noise or electronic sounds). This is achieved by allowing the child to volitionally control which audio plays through their visual attention allocation. The gaze-contingent audio delivery system 300 represents a new approach to measuring auditory social attention that overcomes limitations of prior passive listening paradigms by engaging the child's active attention and measuring sustained preference rather than merely initial orienting responses.

[0099] At the functional level, FIG. 3 depicts the gaze-contingent audio delivery system 300 as an integrated mechanism including a display screen 305 presenting visual target areas that serve as audio selection controls, a gaze detection and tracking component 310 that continuously monitors the child's point of gaze on the display, a real-time processing logic 315 that determines which audio stimulus should be activated based on current gaze location and dwell time, audio control logic 320 implementing decision rules for audio activation, switching, and muting, as well as an audio output mechanism 325 that delivers the selected audio stimulus to the child. Each of these components work in tight temporal coordination to provide responsive audio feedback (typically within 200-500 milliseconds) to the child's gaze shifts. This creates a contingent relationship between visual attention and auditory stimulation that reveals the child's auditory’ preferences through their gaze behavior.Display Screen with Defined Areas of Interest

[0100] The display screen 305 presents visual target areas that function as gaze-operated audio controls during gaze-contingent testing. The display screen 305 corresponds to the display device 130 described in FIG.1, but FIG. 3 specifically illustrates the spatial organization of areas of interest (AOIs) and their functional relationship to audio delivery during auditory social attention tests, namely, the Motherese vs. Highway test 235 and Motherese vs. Techno test 240.Docket No. 1133.124WO1

[0101] The display screen 305 is divided into at least two defined areas of interest: a social AOI 330 and a non-social AOI 335. Both of the social AOI 330 and the non-social AOI 335 are typically, though not necessarily, positioned on opposite sides of the display screen 305, such as left and right halves, to provide clear spatial separation enabling unambiguous determination of which area the child is fixating. In one example, each of the social AOI 330 and the non-social AOI 335 occupies approximately 40-60% of the screen width, with a small neutral zone between them to account for gaze estimation uncertainty near boundaries.

[0102] The social AOI 330 is associated with delivery’ of social auditory stimuli, specifically infant-directed speech or motherese. Within the social AOI 330, a visual stimulus is displayed that indicates or represents the social audio content. In one configuration, the visual stimulus comprises a static or dynamic image of the actress who is speaking in the audio track, providing visual-audio correspondence wherein the child sees the speaker while hearing their voice. In alternative examples, the visual stimulus may comprise a non-face social image, an abstract visual placeholder, or animated graphics that move or change when the corresponding audio is active, providing visual feedback that confirms audio activation. The specific visual content within the social AOI 330 is selected to be engaging to toddlers while not being so interesting as to draw attention based solely on visual properties. Rather, the visual content serves primarily as a spatial marker indicating where to look to activate social audio.

[0103] The non-social AOI 335 is associated with delivery’ of non-social auditory stimuli, specifically traffic sounds (in Motherese vs. Highway test 235) or electronic sounds (in Motherese vs. Techno test 240). Within the non-social AOI 335, a visual stimulus is displayed indicating or representing the non-social audio content. In one implementation, the visual stimulus comprises abstract geometric shapes, patterns, or images conceptually related to the audio (such as images of cars or roads for traffic sounds as well as abstract technological or electronic graphics for techno sounds).Docket No. 1133.124WO1Similar to the social AOI 330, the visual content serves primarily as a spatial marker rather than as a highly engaging visual stimulus.

[0104] An audio active indicator 340 (such as a speaker icon shown in FIG, 3) may be displayed within the social AOI 330 when social audio is playing, thereby providing visual feedback to the child that their gaze has successfully activated the audio. In other examples, the audio active indicator 340 may comprise an animated speaker icon with sound waves, a changing color or brightness, animated elements, or other visual cues.Conversely, an audio inactive indicator 345 (such as a muted speaker icon shown in FIG. 3) is displayed within the non-social AOI 335 when that audio is muted, or the audio active indicator 340 when non-social audio is playing. These visual indicators provide feedback that reinforces the contingent relationship between gaze location and audio delivery, helping even young toddlers understand and engage with the gaze-contingent mechanism.Gaze Detection and Tracking Component

[0105] The gaze detection and tracking component 310 continuously monitors the child's point of gaze on the display screen 305 throughout the test duration, providing the real-time gaze data necessary for implementing contingent audio control. The eye-tracking device 125 shown in FIG. 1 monitors the child's eye 350, which captures images of the eye at high temporal frequency, typically 60 to 600 Hz, or in some examples at 120 Hz. The eye -tracking device 125 identifies pupil position and corneal reflection position in each captured eye image and uses established geometric models to calculate gaze direction and point-of-gaze on the display screen 305.

[0106] A gaze vector 355 represents the calculated direction from the child's eye 350 to their point of fixation on the display screen 305. The gaze vector 355 is typically represented as an angle or as a directional vector in three-dimensional space. For purposes of the gaze-contingent implementation, however, the gaze vector 355 is ultimately projected onto the two-dimensional display screen 305 to determine x-y screen coordinates.Docket No. 1133.124WO1

[0107] A fixation point 360 represents the calculated x-y coordinates on the display screen 305 where the child is currently looking. The fixation point 360 is updated at the sampling rate of the eye-tracking system (typically 120 times per second), generating a continuous stream of gaze coordinate data. Not every gaze coordinate corresponds to a true fixation in the oculomotor sense. Rapid sequences of coordinates during saccadic eye movements or during blinks represent transition periods rather than stable fixations.Therefore, the gaze detection component 310 applies filtering to identify stable fixations, typically defined as maintaining gaze within a spatial threshold. In some examples, this spatial threshold is 2 degrees of visual angle, or approximately 30-50 pixels on a typical display at typical viewing distance for a minimum duration, which can be for at least 100 milliseconds.

[0108] The real-time processing logic 315 performs several operations on the incoming gaze data stream to support gaze-contingent audio control. First, gaze coordinates are captured at each sampling instant, generating a stream of x-y coordinate pairs with associated timestamps. Second, AOI boundary detection determines whether each fixation point 360 falls within the social AOI 330, the non-social AOI 335, or a neutral zone between them. Third, dwell time calculation tracks how long the child's gaze remains continuously within a particular AOI, accumulating fixation duration as long as the fixation point 360 stays within that AOI's boundaries. Fourth, threshold comparison evaluates whether the accumulated dwell time within an AOI has exceeded a minimum duration threshold required to trigger audio activation, which in some examples is typically 100 to 500 milliseconds. This minimum duration threshold prevents spurious audio switching due to brief glances or gaze estimation noise. Audio is only activated when the child sustains fixation within an AOI for at least the minimum threshold duration, indicating intentional attention rather than transient gaze drift.

[0109] The real-time processing logic 315 operates continuously throughout the test duration, updating AOI occupancy and dwell time calculations at each sampling instant. This continuous monitoring enablesDocket No. 1133.124WO1rapid detection of gaze location changes and correspondingly rapid audio switching, providing responsive feedback to the child's attention shifts.Audio Control Logic

[0110] The audio control logic 320 implements decision rules that determine, based on the real-time gaze data, which audio stimulus should be delivered at each moment during the test, Tire audio control logic 320 operates as a state machine that transitions between audio delivery states based on gaze location and dwell time.

[0111] As shown in FIG. 3, a first decision point 365 evaluates whether the child's gaze has entered the social AOI 330. In some examples, this decision is made by comparing the current fixation point 360 to the spatial boundaries defining the social AOI 330. If the fixation point 360 falls within the social AOI 330 boundaries, the decision is “YES” and processing proceeds to evaluate dwell time. If the fixation point 360 falls outside the social AOI 330 (either in the non-social AOI 335, in a neutral zone, or off-screen), then the decision returns “NO” and social audio remains muted.

[0112] A second decision point 370 evaluates whether the dwell time within the social AOI 330 has met or exceeded a dwell time threshold. In some implementations, the dwell time threshold is set to a value between 100 and 500 milliseconds, and in some examples 300 milliseconds. This dwell time threshold is selected based on several technical considerations: (i) it must be long enough to distinguish intentional fixation from transient gaze drift or eye-tracking noise, thereby reducing spurious audio switching that would confuse the child; (ii) it must be short enough to provide responsive feedback that the child perceives as contingent on their gaze behavior, as latencies exceeding approximately 500 milliseconds may not be perceived as contingent by toddlers; and (iii) it should be short enough to enable multiple audio switches during the test duration, allowing measurement of exploration behavior. If the dwell time meets or exceeds the dwell time threshold (“YES”), processing proceeds to a social audio activation component 375. If dwell time threshold has not yet been reached (“NO”),Docket No. 1133.124WO1then social audio remains muted and the social attention assessment system 100 continues to accumulate dwell time as long as gaze remains within the social AOI 330.

[0113] The social audio activation component 375 triggers playback of the social audio stimulus through the audio output system 135. Once activated, the social audio continues to play as long as the child's gaze remains within the social AOI 330. If gaze exits the social AOI 330, such as by shifting to the non-social AOI 335 or looking away from the screen, the social audio activation component 375 mutes or pauses the social audio. In some examples, this occurs with a brief fade-out (such as between 50 to 100 milliseconds) to avoid abrupt audio cutoffs that might startle the child.

[0114] A first audio muting component 376 and a second audio muting component 378 implement the complementary state wherein audio is not delivered. The first audio muting component 376 is activated when the “NO” path from the first decision point 365 indicates that gaze is not within the social AOI 330. The second audio muting component 378 is activated when the “NO” path from the second decision point 370 indicates that gaze is within the social AOI 330 but dwell time has not yet met the dwell time threshold. The first audio muting component 376 and the second audio muting component 378 ensure that audio is only delivered when the child demonstrates sustained attention to the designated AOI, implementing the contingency that is central to the gaze-contingent paradigm.

[0115] In those implementations where non-social audio is actively delivered (as opposed to silence or muting), the audio control logic 320 implements parallel decision logic for the non-social AOI 335. When gaze enters the non-social AOI 335 and dwell time exceeds the dwell time threshold, non-social audio (such as traffic sounds or techno sounds) is activated, while social audio is muted. The social attention assessment system 100 thus alternates between three possible states: (i) social audio playing while gaze dwells in social AOI 330; (ii) non-social audio playing while gaze dwells in non-social AOI 335; and (iii) silence or ambientDocket No. 1133.124WO1background when gaze is in neutral zones, during saccades between AOIs, or when the child looks away from the screen.Audio Output Mechanism

[0116] The audio output mechanism 325 delivers the selected audio stimuli to the child through acoustic transducers. In some examples, these acoustic transducers are audio speakers positioned adjacent to or integrated with the display device 130. A social audio component 380 of the audio output mechanism 325 includes audio recordings of infant-directed speech, song, or social vocalizations. For the Motherese vs. Highway test 235 and the Motherese vs. Techno test 240, the social audio component 380 specifically includes an actress telling a story, such as a narrative about a teddy bear or other child-friendly topic. In some examples the actress uses motherese prosody characterized by elevated pitch (typically 200-400 Hz fundamental frequency compared to 100-200 Hz for adult-directed speech), exaggerated pitch excursions, slower tempo, repetition of key phrases, high frequency of question intonation, and emotional expressiveness. The actress is recorded in a quiet environment with high-quality audio equipment to ensure clear, artifact-free speech. The audio duration matches the test duration, and the story content is structured to maintain continuous speech without long pauses that might confuse the child.

[0117] A non-social audio component 385 includes audio recordings of non-human, non-communicative sounds selected to provide appropriate control stimuli. For the Motherese vs. Highway test 235, the non-social audio comprises traffic sounds including recordings of vehicle engines (such as, by way of example, cars, trucks, motorcycles), horn honks, vehicles accelerating and decelerating, and ambient road noise. Traffic sounds are selected because they provide complex, temporally varying auditory stimulation with amplitude modulation and spectral content somewhat similar to speech (both contain broadband noise and temporal structure), yet lack any human vocal or communicative properties. For the Motherese vs. Techno test 240, the non-social audio includes electronic or technological sounds including synthesized musical tones, computer-generated beeps andDocket No. 1133.124WO1bloops, digital sound effects, or abstract electronic music lacking vocals. Techno sounds provide an alternative non-social control that differs from traffic sounds, enabling convergent assessment of speech preference across different non-social comparison conditions.

[0118] In some configurations, the non-social audio component 385 is muted rather than actively played. This means that when the child looks at the non-social AOI 335, the result is silence (in other words, the cessation of social audio) rather than active presentation of traffic or techno sounds. This muting approach simplifies the paradigm from the child's perspective: looking at the social side produces motherese, looking away stops the motherese. In alternative configurations, the non-social audio component 385 is actively presented, such that the child experiences a choice between two different sound types. However, it should be noted that active non-social audio presentations may reduce preference effects because some children may find traffic or techno sounds interesting in their own right.

[0119] The audio output mechanism 325 also includes a volume control 390 that maintains the audio output at appropriate levels for toddler testing. The social attention assessment system 100 calibrates the audio volume to be clearly audible at the testing distance (typically 60-70 cm) without being uncomfortably loud. In some implementations, this is generally at 60-75 dB sound pressure level. Volume is tested prior to each session and may be adjusted based on ambient noise levels in the testing environment.Consistent volume across children is important for standardization, ensuring that differences in audio preference reflect social attention differences rather than volume perception differences.Continuous Monitoring Loop and System Feedback

[0120] As noted in FIG. 3, a continuous monitoring loop provides ongoing integration between the display screen 305, the gaze detection component 310, the audio control logic 320, and the audio output mechanism 325 throughout a test duration. This feedback pathway, illustrated by the dashed line in FIG. 3, represents the continuous cycle of operations: (i) gazeDocket No. 1133.124WO1detection captures current gaze coordinates; (ii) real-time processing determines AOI occupancy and dwell time; (iii) the audio control logic 320 makes activation / muting decisions; (iv) the audio output mechanism 325 delivers the selected stimulus; (v) the child perceives the audio and may adjust their gaze based on audio preference; and (vi) the gaze detection component 310 captures the updated gaze position, thus continuing the cycle.

[0121] This closed-loop operation continues throughout the entire test duration. As noted above, in some examples this test duration is 61 seconds for the Motherese vs. Highway 235 test and 58 seconds for the Motherese vs. Techno 240 test. This closed-loop operation generates a detailed record of the temporal dynamics of audio preference. The social attention assessment system 100 logs, at each sampling instant or at a reduced logging frequency (such as 10 to 30 Hz), the current gaze coordinates, the current AOI occupancy, the current audio state (social active, non-social active, or muted), and cumulative duration spent triggering each audio type. This logged data enables calculation of the primary metric, which is the percent fixation on social AOI triggering motherese audio. This logged data also facilitates the computation of secondary metrics such as number of audio switches, mean listening bout duration, and temporal patterns of preference.Timing Specifications and Technical Parameters

[0122] The gaze-contingent audio delivery’ system 300 operates according to specific timing specifications that have been empirically optimized for toddler assessment. These timing parameters balance competing objectives of responsiveness, stability, and measurement validity.

[0123] One of these timing parameters is the dwell time threshold, as discussed above. In some examples, the dwell time threshold is set to a value between 300 and 500 milliseconds. In one implementation the dwell time threshold is approximately 300 milliseconds (ms). The lower bound of this range, or 300 ms, was selected to exceed the typical duration of infantDocket No. 1133.124WO1saccades (rapid eye movements between fixations, which typically last 30-80 ms) and brief fixations during visual scanning, ensuring that audio activation reflects sustained attention rather than transient gaze drift. If the dwell time threshold were set too low (such as below 200 ms), spurious audio switching would occur frequently due to normal exploratory eye movements, degrading the child's experience and reducing measurement validity. The upper bound of this range, or 500 ms, was selected to ensure responsive feedback. If the upper bound of the dwell time threshold were set too high (such as above 500-600 ms), the child might not perceive the contingency between their gaze and audio delivery’, reducing engagement with the paradigm and potentially leading the child to look away from the screen entirely.

[0124] Another one of the timing parameters is the sampling rate for gaze detection. This sampling rate, in some examples, is typically 60 to 120 Hz, depending on the eye-tracking hardware employed. A minimum sampling rate of 60 Hz (or one sample every 16.7 ms) is typically needed to provide sufficient temporal resolution to detect dwell time thresholds in the 300-500 ms range with adequate precision. Higher sampling rates (such as 600 Hz available on some commercial eye-tracking systems) provide improved temporal precision and enable more sophisticated filtering of eye-tracking noise but increase data volume and processing demands. In some examples of the gaze-contingent audio deliv ery system 300, sampling at 120 Hz (one sample every 8.3 ms) provides a practical balance between temporal resolution and computational efficiency.

[0125] Another timing parameter is the audio switching latency, which is defined as the time from detection of gaze entry into a new AOI (meeting the dwell time threshold) to onset of the corresponding audio stream. In some configurations, the social attention assessment system 100 maintains this audio switching latency below 200 to 500 milliseconds. This rapid switching is achieved through: (i) continuous gaze monitoring rather than periodic checking, enabling immediate detection of AOI entry; (ii) pre-loading both audio streams into memory buffers rather than loading on-demand,Docket No. 1133.124WO1eliminating file access delays; (iii) efficient audio control logic implemented in compiled code rather than interpreted scripts; and (iv) use of low-latency audio hardware and drivers. The technical effect of achieving low audio switching latency is that toddlers perceive clear contingency between their gaze and audio delivery, maintaining engagement with the paradigm and enabling valid measurement of sustained audio preference.Operation of the Gaze-Contingent Audio Delivery System

[0126] In operation during administration of an auditory’ social attention test, the gaze-contingent audio delivery / system 300 functions as follows. Prior to beginning the gaze-contingent test, the social attention assessment system 100 implements a training procedure (such as the Dog-Cat Training paradigm described above in relation to FIG. 2) to familiarize the child with the gaze-contingent mechanism. During training, the child learns that their gaze location controls what they hear, establishing the contingency concept in a simplified context. For example, the child looks at a dog image and hears a barking sound, and the child looks at a cat and hear a meowing sound. Once training is completed, the social attention assessment system 100 proceeds to the actual test with speech and non-social sounds,

[0127] The gaze-contingent test begins with the display screen 305 shows the social AOI 330 and the non-social AOI 335 with their associated visual markers. Both audio streams, such as the social audio component 380 and the non-social audio component 385, are loaded into memory buffers and ready for instant playback. The social attention assessment system 100 enters an initial neutral state wherein no audio plays, waiting for the child's first sustained fixation.

[0128] As the child looks at the display screen 305, the gaze detection component 310 continuously captures gaze coordinates from the child's eye 350, calculates the gaze vector 355, and determines the fixation point 360 on the display screen 305. When the fixation point 360 enters the social AOI 330, the real-time processing logic 315 detects this AOI entry and begins accumulating dwell time. If the child maintains gaze within the social AOIDocket No. 1133.124WO1330 for the dwell time threshold duration (typically 300 ms), then the audio control logic 320 evaluates the first decision point 365 (gaze in social AOI? “YES”) and the second decision point 370 (dwell time met dwell time threshold? “YES”), triggering the social audio activation component 375 to begin playback of the social audio component 380.

[0129] The child now hears the actress speaking with motherese prosody. The audio output mechanism 325 delivers this speech through speakers at calibrated volume. Typically developing toddlers find motherese highly engaging and preferable to non-social sounds, and therefore often maintain fixation within the social AOI 330 for extended durations (often 5-20 seconds or longer) to continue hearing the speech. During this sustained attention, the gaze detection component 310 continues monitoring gaze, the real-time processing logic 315 confirms continued AOI occupancy, and the social audio activation component 375 maintains playback of the social audio component 380.

[0130] At some point, the child's gaze may shift away from the social AOI 330. This gaze shift may be motivated by exploration (curiosity about the other side of the display screen 305), by waning interest in the speech, by¬ distraction, or by other factors. When the fixation point 360 exits the social AOI 330, the real-time processing logic 315 detects this AOI exit. The audio control logic 320 evaluates the first decision point 365, which now returns “MO” (gaze not in social AOI). This triggers the first audio muting component 376 to stop or mute the playback of the social audio component 380. This audio cessation typically occurs within 100-200 milliseconds of gaze exit, providing clear feedback that looking away stops the audio.

[0131] If the child's gaze shifts to the non-social AOI 335, the same sequence occurs: dwell time is accumulated, and if the dwell time threshold is met, playback of the non-social audio component 385 is initiated, assuming that non-social audio presentation is turned on. In other examples that employ muted non-social audio, the result is simply continued silence.Docket No. 1133.124WO1

[0132] The child may then shift gaze back to the social AOI 330, re¬ triggering playback of the social audio component 380 after meeting the dwell time threshold again. This cycle of looking to activate audio, listening while maintaining gaze, looking away to stop audio, and optionally looking back to re-activate audio may repeat multiple times during the test duration. The number of such cycles, and the cumulative duration spent triggering social audio versus non-social audio, provides the quantitative metrics that characterize auditory social attention preference.

[0133] Throughout this process, the continuous monitoring loop of the gaze-contingent audio delivery’ system 300 maintains tight integration between the display screen 305, the gaze detection component 310, the audio control logic 320, and the audio output mechanism 325, ensuring that the audio delivered at each moment accurately reflects the child's current gaze location and attention state. In some examples, the social attention assessment system 100 logs all gaze coordinates, AOI occupancy states, and audio delivery' states throughout the test, providing a complete record for post-hoc metric extraction and quality’ control assessment.Alternative Embodiments and Optional Features

[0134] Several alternative configurations and variations of the gazecontingent audio delivery' system 300 may be implemented while retaining the core principle of gaze-controlled audio selection. In alternative examples, more than two AOIs are employed, such as three or four distinct regions each associated with different audio stimuli. For example, a three- AOI implementation can present motherese on the left, traffic sounds in the center, and techno sounds on the right, enabling direct comparison of motherese preference relative to multiple non-social controls within a single test. It should be noted, however, that increasing the number of AOIs reduces the size of each AOI and may’ increase task complexity’ beyond the cognitive capabilities of very young toddlers (12-18 months). Especially for this age range, the social attention assessment system 100 typically uses the two-AOI implementations.Docket No. 1133.124WO1

[0135] In another alternative examples, the visual stimuli displayed within the AOIs are dynamic rather than static, such as showing a video of the actress speaking (video synchronized with audio) in the social AOI 330, or showing animated geometric patterns in the non-social AOI 335. Dynamic visual content may enhance engagement but introduces potential confounds wherein visual interest in the dynamic content, independent of audio preference, influences gaze allocation. Thus, the social attention assessment system 100 typically uses static or minimally dynamic visual content to minimize such confounds and ensure that measured preferences primarily reflect auditory rather than visual interest.

[0136] The dwell time threshold may be varied across some examples or even adaptively within a single test. By way of example and not limitation, a shorter dwell time threshold (such as between 100-200 ms) provides more responsive feedback and enables children to trigger audio activation more easily. This potentially improves engagement with very young or less attentive children, but may increase spurious switching. A longer dwell time threshold (such as between 400-500 ms) reduces spurious switching and ensures that audio activation reflects sustained intentional attention. This long dwell time threshold, however, may reduce responsiveness to the point that some children do not perceive contingency. The social attention assessment system 100 typically uses a dwell time threshold of approximately 300 ms to balance these considerations. This dwell time threshold is also based on empirical testing with over 700 toddlers in each of the two auditory’ tests.

[0137] In some configurations, hysteresis can be implemented in the audio control logic 320 to prevent rapid audio oscillation when gaze dwells near AOI boundaries. For example, once playback of the social audio component 380 is activated by meeting or exceeding the dwell time threshold within social AOI 330, an exit threshold that is shorter than the dwell time threshold (such as the exit threshold being equal to about 100 ms outside the AOI) might be required to deactivate playback of the social audio component 380. Thus, the dwell time within the social AOI 330 required toDocket No. 1133.124WO1activiate playback of the social audio component 380 is longer than the dwell time outside of the social AOI 330 required to deactivate playback of the social audio component 380, This hysteresis, also known as a deadband or switching delay asymmetry, stabilizes audio delivery when gaze hovers near boundaries.

[0138] In other examples, the spatial arrangement of AOIs may be varied, such as positioning the social AOI 330 on the left for some children and on the right for others, to control for potential spatial biases. For example, some children may have inherent left-side or right-side gaze preferences. In a typical implementation, examples of the social attention assessment system 100 counterbalance the AOI positions across children. This means that approximately half of the children receive social audio on the left and half receive it on the right,

[0139] Visual feedback indicators (speaker icons, color changes, animations) may be enhanced, minimized, or eliminated in some examples, depending on the desired balance between providing clear feedback to the child and avoiding visual confounds. In some implementations, no visual indicators are used beyond the static visual markers defining each AOI location. In other configurations, salient animated indicators provide strong visual feedback reinforcing the contingency.

[0140] The gaze-contingent audio delivery system 300 may be applied to tests beyond the auditory domain. For example, a gaze-contingent visual test might present static images that trigger dynamic video playback when fixated, or might use gaze location to control which of two simultaneously playing videos has audio enabled. Such extensions maintain the core principle of measuring preference through volitional gaze-based control.

[0141] Those skilled in the art will recognize that features of the gazecontingent audio delivery' system 300 may be combined with features from other system components or modified in various ways without departing from the inventive concept. By way of example and not limitation, the specific social audio content (motherese) could be replaced with orDocket No. 1133.124WO1supplemented by other social sounds such as infant laughter, singing of familiar songs, other children’s voices, or playful vocalizations. The non¬ social audio could include other categories such as white noise, pure tones, animal sounds, environmental sounds (rain, wind), or instrumental music. The visual markers within AOIs could be enhanced with more elaborate graphics, videos, or animations, or simplified to minimal placeholders.

[0142] The two-AOI spatial arrangement could be modified to vertical (top-bottom) rather than horizontal (left-right) positioning, to quadrants rather than halves, or to central versus peripheral arrangements. The AOI boundaries could be made visible to the child (e.g., with colored borders) or kept invisible.

[0143] The dwell time threshold could be made adaptive, such as starting with a low threshold (e.g., 200 ms) to help the child discover the contingency, then increasing to a higher threshold (e.g., 400 ms) for more stringent measurement of sustained preference. Multiple thresholds could be applied simultaneously, logging different metrics at different dwell durations (e.g., "short fixations" 200-500 ms, "sustained fixations" >500 ms). Tire gaze-contingent principle could be extended beyond audio control to control other stimulus dimensions such as video playback (gaze activates a social video vs. non-social video), visual contrast or brightness (gaze enhances visibility of preferred stimulus type), stimulus size or position, or even haptic feedback in embodiments incorporating touch -based components.

[0144] Those skilled in the art will appreciate that many modifications and equivalents are possible within the scope of the gaze-contingent audio delivery system 300. The specific implementation details, timing parameters, stimulus types, and AOI arrangements may be varied while maintaining the fundamental principle of measuring auditory’ social attention through gaze-controlled audio selection. The gaze-contingent audio delivery system 300 described with reference to FIG. 3 should be understood as illustrative of the principles of gaze-contingent assessment rather than limiting of its scope,Docket No. 1133.124WO1with the full breadth of protection defined by the appended claims and their equivalents.Normative Reference Database

[0145] FIG. 4 illustrates the architecture and organization of a normative reference database 400 that facilitates percentile-based social attention profiling. The normative reference database 400 shown in FIG. 4 illustrates specific examples of the normative database 180 described in FIG.l. The normative reference database 400 stores validated eye-tracking data from a large population of neurotypically developing children, organized with systematic stratification by age and gender to enable precise developmental comparisons when assessing individual children. The normative reference database 400 provides continuous quantitative scores indicating where a child's social attention capabilities fall relative to age-matched and gender- matched peers rather than binary determinations of typical versus atypical development.

[0146] At the architectural level, FIG. 4 depicts the architecture of the normative reference database 400 as a structured data repository. This structured data repository includes database characteristics 410 defining the scope and composition of the reference population, a stratification structure 420 organizing children into developmental age bins with separate statistical reference values for each bin, gender stratification within each age bin maintaining separate normative values for male and female children, an individual record structure 430 defining data elements stored for each child in the database, computed statistical reference values 440 derived from the population data, and mechanisms for clinical application 450 enabling age¬ gender matched comparison of new patients to the appropriate normative reference group. This hierarchical organization enables efficient retrieval of relevant comparison data while maintaining the statistical precision necessary for accurate percentile ranking across the wide developmental range (12-48 months) encompassed by the target population.Docket No. 1133.124WO1Database Characteristics and Composition

[0147] The database characteristics 410 define the scope, size, and composition of the normative reference population, establishing the foundation for statistical validity of percentile rankings derived from the normative reference database 400. In the examples shown in FIG. 4, the total sample size comprises at least 2,000 children who have completed one or more of the six eye-tracking tests described in FIG. 2. This large sample size is critical for several reasons: (i) it provides sufficient statistical power to detect true developmental patterns while minimizing effects of individual variability; (ii) it enables calculation of stable percentile distributions, particularly at the tails (e.g., 5th and 95th percentiles) where fewer data points exist; (iii) it permits stratification into multiple age bins and gender groups while maintaining adequate sample sizes within each stratum (typically at least 200-400 children per age-gender stratum); and (iv) it represents diverse demographic, ethnic, and socioeconomic backgrounds, enhancing generalizability to the broader population of children. The lower bound of 2,000 children represents a minimum sample size adequate for these purposes based on statistical power analyses. The nonnative reference database 400 will continue to grow over time as additional children are assessed and their data incorporated, as described further below,

[0148] The age range covered by the normative reference database 400 spans 12 to 48 months, corresponding to the validated age range for administration of the multi-domain eye-tracking test battery 200. The lower bound of 12 months represents the youngest age at which the eye-tracking tests have been systematically validated, though some children as young as 9 or 10 months may successfully complete testing. Below 12 months, toddlers typically have insufficient attention span and cooperation to complete the test battery reliably. The upper bound of 48 months represents the oldest age in the current validation sample, though the tests could potentially be administered to older children. Above 48 months, children are typically assessed using different developmental evaluation tools appropriate for preschool and early school-age populations.Docket No. 1133.124WO1

[0149] The neurotypical development criterion specifies that children included in the normative database are screened to exclude those with known developmental delays, diagnosed autism spectrum disorder, or other conditions that might affect social attention development. Specifically, children typically are included in the normative reference database 400 if they: (i) score within normal limits (within 1 standard deviation of age-expected means, or more specifically within 1.5 standard deviations) on standardized developmental assessments such as the Mullen Scales of Early Learning; (ii) do not meet diagnostic criteria for autism spectrum disorder based on assessments such as the Autism Diagnostic Observation Schedule (ADOS-2) administered by licensed clinical psychologists; (iii) do not have diagnosed neurological, sensory, or genetic conditions known to affect development; and (iv) demonstrate typical adaptive functioning on measures such as the Vineland Adaptive Behavior Scales. This strict inclusion criterion ensures that the normative reference database 400 represents truly typical development, providing an appropriate reference for identifying children with atypical social attention patterns,

[0150] Multi-site collection indicates that data in the normative reference database 400 has been collected at multiple testing sites, including multiple pediatric clinics, research laboratories, and preschool settings, rather than from a single institution. In some examples, data has been collected from at least 3 sites. In one example, data was collected from approximately 37 pediatric office locations and 2 research laboratory locations. Multi-site collection enhances generalizability by ensuring that nonnative values are not specific to local population characteristics, testing environments, or administrator techniques at a single site, instead representing performance across diverse settings that reflect real-world application contexts.

[0151] Standardized protocol ensures that all children in the normative reference database 400 were assessed using consistent methodology.Standardization includes: (i) use of identical eye-tracking hardware and software across sites; (ii) identical test stimuli and presentation parameters;Docket No. 1133.124WO1(iii) standardized calibration procedures with defined acceptance criteria; (iv) consistent administration instructions provided to parents and children; (v) standardized data quality criteria for inclusion / exclusion of test sessions; and (vi) training and certification of technicians administering tests to ensure inter-rater reliability. The technical effect of having this standardized protocol is that variability in the normative reference database 400 reflects true developmental differences between children rather than methodological differences between testing sessions or sites.

[0152] Quality control indicates that data in the normative reference database 400 has undergone systematic quality assessment, with exclusion of test sessions failing to meet minimum quality criteria. Quality control procedures are described in greater detail in the context of the individual record structure 430 below, but generally include exclusion of sessions with poor calibration, insufficient looking time, excessive tracking loss, or other indicators of invalid data. In some implementations, approximately 10-20% of collected test sessions are excluded based on quality criteria. This ensures that the normative reference database 400 contains only high-quality, reliable data suitable for comparison purposes.Stratification Structure

[0153] The stratification structure 420 organizes children in the normative reference database 400 into age bins, also referred to as age groups, age strata, or age brackets. This is where children falling within a defined age range are grouped together for statistical analysis. This stratification is useful because social attention capabilities change rapidly during the toddler period. A 12-month-old exhibits substantially different eye-tracking patterns compared to a 24-month-old or 36-month-old, reflecting ongoing brain development and social learning. By stratifying by age, the normative reference database 400 enables age-matched comparisons wherein an individual child is compared only to other children of similar age, accounting for these developmental changes.Docket No. 1133.124WO1

[0154] In some examples, age bins are defined at 4-month intervals, generating distinct age groups spanning the 12-48 month range. The 4-month interval is selected based on empirical analysis indicating that this interval captures meaningful developmental changes while maintaining adequate sample sizes within each bin. Smaller intervals (such as, by way of example, 2 -month or 1 -month bins) would provide finer developmental resolution but would divide the sample into too many small groups, thereby reducing statistical stability. Larger intervals (such as, for example, 6-month or 12-month bins) would increase sample sizes but would group together children at meaningfully different developmental stages, thereby reducing precision of age-matching. The 4-month interval represents an optimized balance validated through analysis showing that within-bin developmental variation is substantially smaller than between-bin variation.

[0155] The examples shown in FIG. 4 illustrate three representative age groups from the complete set of age strata. It should be understood, however, that the actual database implementation includes age bins spanning the entire 12-48 month range, typically yielding 9 age bins, or more generally between 6 and 12 age bins, depending on the specific binning scheme employed.

[0156] Age group 1 (455) of the stratification structure 420 represents the youngest age stratum, spanning 12 to 15 months. Children in age group 1 (455) are in early toddlerhood, typically beginning to walk independently, using first words, and showing emerging joint attention capabilities. Eye¬ tracking patterns at this age show relatively low levels of sustained attention (total looking times often 40-60% of test duration), frequent looking away from the screen, and high variability across children reflecting individual differences in attention maturation. The sample size for age group 1 (455) typically includes 200 to 400 children, or in one example approximately 300 children, distributed across male and female subsets.

[0157] Within age group 1 (455), a first male subset 456 includes male children aged 12-15 months who have completed one or more eye-trackingDocket No. 1133.124WO1tests. The sample size for the first male subset 456 is typically 100 to 200 children, or in one example approximately 150 children. Similarly, a first female subset 458 includes female children in the same age range, with comparable sample size. Gender stratification within each age group is implemented because validation data indicates that male and female toddlers show statistically different social attention patterns on some metrics. For examples, female toddlers showing slightly higher attention to faces and slightly more joint attention cycles compared to male toddlers. These gender differences are subtle (effect sizes typically 0.2-0.4 standard deviations) but systematic, justifying separate normative values for males and females to improve precision of percentile ranking.

[0158] For each age-gender stratum (e.g., males aged 12-15 months), the normative reference database 400 stores computed statistical values for each metric of each test including: mean (average value across all children in that stratum), standard deviation (measure of variability across children), and percentile ranks (values corresponding to the 5th, 10th, 25th, 50th, 75th, 90th, and 95th percentiles of the distribution). These statistical summaries enable efficient comparison of an individual child's metrics to the normative distribution without requiring access to raw data from all individual children.

[0159] Age group 2 (460) represents a middle age stratum from 16 to 23 months. In some examples using one binning scheme, age group 2 (460) spans 16 to 19 months, while in other examples using a different binning scheme, the span is 20 to 23 months. The specific range of age group 2 (460) depends on whether 4-month bins are strictly maintained. Children in age group 2 (460) typically show increased attention span compared to age group 1 (455), improved joint attention capabilities, emerging language comprehension, and greater variability in social attention patterns as individual developmental trajectories diverge. The sample size for age group 2 (460) typically includes 400 to 600 children. This reflects the fact that age group 2 (460) encompasses the peak age for autism screening (18 months) and therefore represents a larger proportion of the clinical population undergoing assessment.Docket No. 1133.124WO1

[0160] Within age group 2 (460), a second male subset 462 and a second female subset 464 maintain gender stratification. In some examples, the second male subset 462 and the second female subset 464 each typically include about 200 to 300 children, while in at least one example include approximately 250 children. Statistical reference values (mean, SD, percentiles) are computed separately for each gender within age group 2 (460).

[0161] Age group 3 (465) represents an older age stratum, typically spanning 24 to 48 months. In some examples, age group 3 (465) spans 24 to 27 months in one bin, 28 to 31 months in another bin, and so forth through to 40-48 months in the oldest bin. Children in these older age ranges show substantially more mature attention capabilities, including sustained attention for longer durations, more systematic visual scanning, reliable joint attention following, and stronger language-mediated attention control. The sample size for age group 3 (465), and subsequent older age groups, typically includes 200 to 400 children per 4-month bin. In at least one example the sample size is approximately 300 children.

[0162] Within age group 3 (465), a third male subset 466 and a third female subset 468 maintain gender stratification. Each of the third male subset 466 and the third female subset 468 typically include 100 to 200 children. In some examples, each of the third male subset 466 and the third female subset 468 includes approximately 150 children per 4-month age-gender stratum. Statistical reference values are computed as described for the younger age groups.

[0163] It should be understood that while FIG. 4 illustrates three age groups for clarity, the actual normative reference database 400 implementation includes age groups covering the complete 12-48 month range, typically yielding age groups at intervals such as: 12-15 months, 16-19 months, 20-23 months, 24-27 months, 28-31 months, 32-35 months, 36-39 months, and 40-48 months. In some configurations, the final age group (40-48 months) spans a wider interval (9 months rather than 4 months)Docket No. 1133.124WO1because: (i) developmental changes are less rapid in older toddlers compared to younger toddlers, making wider binning acceptable; (ii) fewer children in the upper age range undergo screening, resulting in smaller sample sizes that benefit from wider binning; and (iii) most clinical applications focus on the 12-36 month age range where early identification is most valuable.Individual Record Structure

[0164] The individual record structure 430 defines the data elements stored for each child in the normative reference database 400, providing the raw material from which statistical reference values are computed. The individual record structure 430 includes demographics 470, test metrics 475, and quality flags 480.

[0165] Demographics 470 includes identifying and descriptive information about each child. By way of example and not limitation, a subject identification (ID) provides a unique identifier for each child. This facilitates linkage of multiple test sessions for children who undergo repeated assessments while protecting privacy through de-identification. It should be noted that the subject ID is a numeric or alphanumeric code that does not contain personally identifying information such as names or birthdates. Demographics 470 in some examples also includes age. Age records the child's chronological age at the time of testing, calculated as months elapsed from birthdate to test date, typically recorded to one decimal place precision (such as 18.3 months). In some implementations, demographics 470 records biological sex as male or female for stratification purposes. Collection site, which is included in some configurations of the demographics 470, identifies the location where data was collected. This facilitates analysis of site effects (where the test(s) was administered) and maintains data quality monitoring. Some versions of the demographic 470 also include the collection date, which records the date of test administration. This data facilitates temporal tracking of database growth and potential analysis of cohort effects if assessment protocols or population characteristics change over time.Docket No. 1133.124WO1

[0166] Test metrics 475 includes the quantitative eye-tracking measurements extracted from each test that the child completed. As described in FIG. 2, in some examples each test generates 9 metrics, yielding a total of up to 54 metrics (9 metrics × 6 tests) for children who complete the full battery, or fewer metrics for children who complete a partial battery. Test metrics 475 can include percent (%) looking time, which indicates the proportion of each test's duration during which valid eye-tracking data was captured. This is typically computed as the time with valid gaze coordinates divided by total test duration. High percent looking time (e.g., >70%) indicates good engagement and cooperation, while low percent looking time (e.g., <40%) may indicate fussiness, inattention, or technical problems. Percent looking time serves as a primary data quality indicator.

[0167] Fixation count indicates the total number of discrete fixation events identified during each test. A fixation event is defined as maintaining gaze within a defined spatial threshold (typically 2 degrees of visual angle, or approximately 30-50 pixels) for at least a minimum duration (typically 100 ms). Fixation count reflects attentional engagement and provides information about visual scanning patterns. Some examples of the test metrics 475 includes mean fixation duration, which represents the average duration of all fixation events during a test. Mean fixation count is typically calculated by summing all fixation durations and dividing by fixation count. In some implementations, the test metrics 475 also include mean fixation duration. Mean fixation duration reflects depth of processing, with longer fixations potentially indicating greater cognitive engagement with stimuli. Typical mean fixation duration values for toddlers range from 200 to 800 ms, with substantial variability across children and across test types. In some configurations, the test metrics 475 include a median fixation duration, which provides an alternative measure of central tendency that is less sensitive to outlier fixations than the mean fixation duration. For distributions of fixation durations that are positively skewed (a few very long fixations pulling the mean upward), the median fixation duration provides a more representative measure of typical fixation duration.Docket No. 1133.124WO1

[0168] Quality flags 480 are indicators of data quality for each test session, enabling systematic quality control and providing the basis for inclusion / exclusion decisions. The several types of quality flags 480 are as follows, with one or more of each of them in any combination being possible in any given implementation.

[0169] Calibration quality indicates the accuracy and precision achieved during eye-tracker calibration prior to testing. Calibration quality is typically quantified as mean angular error across calibration points, measured in degrees of visual angle. Acceptable calibration quality requires accuracy better than approximately 1 to 2 degrees of visual angle. Tests administered following poor calibration (such as accuracy worse than 2 degrees) are excluded from the normative database as gaze coordinates may be unreliable.

[0170] Data completeness is the proportion of test duration for which valid gaze data was captured. Data completeness is closely related to the percent looking time discussed above, but specifically quantifies the absence of tracking loss periods. High data completeness (>80%) indicates that the eyetracking device 125 successfully tracked the child's gaze throughout most of the test, while low data completeness may result from the child moving out of the tracking range, closing eyes, or technical tracking failures,

[0171] Tracking loss percentage (%) quantifies the proportion of test duration during which the eye-tracking device 125 could not detect the child’s gaze. This is calculated as periods with no valid gaze coordinates divided by total test duration. Tracking loss occurs when the child moves their head outside the trackable range, when eyes are closed or partially occluded, when lighting conditions interfere with pupil detection, or when other technical factors prevent gaze estimation. In some examples, an acceptable tracking loss percentage is typically less than between 10% to 30%, with higher tracking loss rates indicating potentially unreliable data.

[0172] Valid test flag provides a binary indicator (such as “valid / invalid”) summarizing whether each test session meets overall quality criteria forDocket No. 1133.124WO1inclusion in analyses. A test is flagged as valid if: (i) calibration quality is acceptable; (ii) percent looking time exceeds a minimum threshold (typically 50%); (iii) tracking loss is below a maximum threshold (typically 30%); and (iv) the technician rating (based on observation of child cooperation and testing conditions) is "good" or "moderate" rather than "poor." Tests flagged as invalid are stored in the database for completeness but are excluded from calculation of normative statistics.[01731 Inclusion status indicates whether each child's data has been incorporated into the normative statistical reference values or has been excluded based on quality or eligibility criteria. Children may be excluded from the normative reference database 400 if: (i) they subsequently receive a diagnosis of autism or developmental delay (even if they appeared typical at the time of testing); (ii) too few valid tests were completed to provide reliable characterization; (iii) demographic information is incomplete; or (iv) participation occurred during a pilot phase before protocols were fully standardized. Inclusion status enables the database to contain comprehensive records for all assessed children while maintaining strict quality control over which children contribute to normative values.Statistical Reference Values

[0174] The statistical reference values 440 include the computed statistical summaries that enable percentile ranking of individual children relative to the normative distribution. As shown in FIG. 4, the statistical reference values 440 include computed for each metric 485 and clinical application 490. Computed for each metric 485 are statistical calculations that are performed separately for each of the eye-tracking metrics, and for each test, within each age-gender stratum. For example, for male children aged 12-15 months who completed the GeoPref test 215, statistical values are computed for each of the 9 metrics extracted from that test (percent fixation social, percent fixation non-social, saccades per second social, etc.). This yields a total of 9 sets of statistical values for that age-gender-test combination. Across all age-gender strata and all tests, the complete set of statistical reference values comprises hundreds or thousands of computed statistics,Docket No. 1133.124WO1organized systematically for efficient retrieval. The computed for each metric 485 may include any of the following, with one or more of each of them in any combination being possible for any given implementation.

[0175] The mean (p) represents the arithmetic average value of each metric across all children in the relevant age-gender stratum who completed the relevant test with valid data. The mean provides a central tendency estimate indicating typical performance for that age-gender group. For example, the mean percent fixation on social images for female children aged 20-23 months on the GeoPref test 215 might be 65%, indicating that on average, girls in this age range spend approximately 65% of viewing time looking at social images, and correspondingly 35% looking at geometric images.

[0176] The Standard Deviation (o) quantifies the variability or spread of the distribution around the mean. Standard deviation is calculated using standard statistical formulas and indicates how much individual children differ from the mean. A small standard deviation (such as 5-10%) indicates relatively homogeneous performance across children, -while a large standard deviation (such as 20-30%) indicates substantial individual differences. For most social attention metrics, standard deviations typically range from 15 to 25%, reflecting considerable individual variability even among neurotypically developing children. The standard deviation is used in calculating z-scores, which are standardized scores indicating how many standard deviations a child's value is from the mean. The standard deviation can also be used to determine whether a child's performance falls within typical ranges.

[0177] Percentile ranks provide the primary reference values used for generating percentile scores in the clinical application. Percentile ranks are calculated by: (i) sorting all children's values for a given metric in ascending order; (ii) determining the value at specific percentage points of the distribution (e.g., the value exceeded by 95% of children and fallen below by 5% of children defines the 5th percentile); and (iii) storing these percentile boundary values. Percentiles are computed at standard intervals including,Docket No. 1133.124WO1but not limited to 5th, 10th, 25th, 50th (median), 75th, 90th, and 95th percentiles. In some examples, finer percentile resolution is computed, such as at 1-percentile intervals (1st, 2nd, 3rd,... 98th, 99th percentiles) to enable precise ranking. When assessing an individual child, their metric value is compared to these percentile boundaries to determine their percentile rank. By way of examples, a value falling between the 25th and 50th percentile boundaries places the child in the 25-50 percentile range, or more precisely at a specific percentile interpolated between those boundaries.

[0178] Z-score thresholds provide an alternative standardized scoring approach based on standard deviation units from the mean. A child's z-score for any metric is calculated as follows: (child's value - mean) / standard deviation. This z-score indicates how many standard deviations above or below the mean the child falls. Z-score thresholds corresponding to specific percentiles (such as z = −1.96 corresponds to approximately the 2.5th percentile for normally distributed data) may be stored and used for classification purposes. Z-scores enable comparison across different metrics that have different scales and units, as all z-scores are expressed in the common unit of standard deviations from the mean.Clinical Application of Normative Reference Database

[0179] The clinical application 490 illustrates how the normative reference database 400 is utilized when assessing an individual child and transforming stored population statistics into child-specific percentile rankings. An individual child assessment 492 represents a new patient, or test subject, undergoing eye-tracking assessment whose data has not previously been in the database. The child completes one or more of the six eye-tracking tests, generating extracted metrics (such as 9 metrics per test), as described previously. The child's age in months and gender are recorded as part of demographic data collection during test setup.

[0180] An age-gender matched comparison 494 includes the process of identifying the appropriate normative reference group and calculating the child's percentile ranks. Specifically, the social attention assessment system 100 first determines which age bin the child falls into based on their age inDocket No. 1133.124WO1months. For example, a 17-month-old child falls into the 16-19 month or 16-23 month age group depending on the binning scheme being used. Second, the social attention assessment system 100 retrieves statistical reference values (mean, SD, percentiles) for that age bin and the child's gender. Then for each metric the child generated, the social attention assessment system 100 compares the child's value to the normative distribution. Finally, the social attention assessment system 100 calculates percentile rank indicating what percentage of age-gender matched children scored at or below the child's value.

[0181] By way of example, if a male child aged 18 months scores 45% fixation on social images in the GeoPref test 215, and the normative data for males aged 16-19 months shows a mean of 65% with SD of 20%, the child's z-score is (45−65) / 20 = −1.0, corresponding to approximately the 16th percentile. This indicates that the child showed less social attention than approximately 84% of age-gender matched peers, potentially flagging reduced visual social attention warranting further evaluation.

[0182] The age-gender matched comparison 494 is repeated for each metric of each test the child completed, generating a comprehensive percentile profile across all assessed social attention capabilities. These percentile ranks are then compiled by the report generator 178 into the clinical assessment report 194, which shows domain-specific summaries and individual test results.Database Operations and Management

[0183] The normative reference database 400 uses, in some examples, the following technical procedures for accessing and maintaining the database. A database query operation called query by age / gender retrieves relevant nonnative values when assessing an individual child. The query specifies the child's age bin and gender, and the normative reference database 400 returns the corresponding statistical reference values for all metrics. Query operations are optimized for rapid retrieval, typically using database indexing on age and gender fields, enabling query responses withinDocket No. 1133.124WO1milliseconds even for databases containing thousands of children. Efficient query performance is important for real-time or near-real-time report generation following test completion.

[0184] A database query operation called retrieve reference statistics specifically extracts the mean, standard deviation, and percentile distributions for the age-gender matched group, providing the values necessary for percentile rank calculation. In some examples, the full set of individual children's raw metric values within the matched group may also be retrieved to enable more sophisticated comparison methods. These methods can include distribution-free percentile ranking or graphical visualization showing the individual child's position within the scatter of normative data points.

[0185] An update / expand dataset query operation represents the capability to incorporate new validated data into the normative reference database 400 over time. As additional children undergo testing and meet inclusion criteria, their data may be added to the normative reference database 400, increasing sample sizes within age-gender strata and potentially improving statistical precision. Database update procedures include appending new individual records to the database, and then recalculating statistical reference values (mean, SD, percentiles) for affected age-gender strata incorporating the new data. The procedure further includes versioning the database to track changes over time, and validating database integrity to ensure consistency and accuracy. In some implementation, database updates occur continuously or at regular intervals (e.g., monthly, quarterly). This provides a "living database" approach wherein the normative reference values continuously improve in precision as sample sizes grow. In other examples, the normative reference database 400 is fixed at a specific version following initial validation, providing stable reference values that do not change over time, which may be preferred for regulatory approval or standardization purposes.

[0186] A database query operation called validate data integrity encompasses procedures for ensuring that the database maintains accuracyDocket No. 1133.124WO1and consistency. Validation procedures include checking for duplicate records that might inflate sample sizes artificially, verifying that age-gender stratifications are correctly assigned, and confirming that statistical calculations are mathematically correct. Validation procedures further include identifying outlier values that might indicate data entry errors, and ensuring that quality-controlled exclusions are correctly implemented such that only valid test sessions contribute to normative values. Data integrity validation is performed periodically (e.g., annually) and whenever significant database updates occur.

[0187] The normative reference database 400, in some examples, uses a relational database management system (RDBMS) such as PostgreSQL, MySQL, Microsoft SQL Server, or similar systems that organize data into structured tables with defined relationships. In the relational structure, separate tables store: (i) individual child demographic records; (ii) test session records linked to child records; (iii) metric values linked to test sessions; (iv) quality flags linked to test sessions; and (v) computed statistical reference values organized by age bin, gender, test, and metric. The relational structure enables efficient querying, flexible reporting, and maintainability as the database grows and evolves. Foreign key relationships between tables ensure referential integrity, preventing orphaned records or inconsistent linkages,

[0188] In other alternative examples, the normative reference database 400 is implemented using other database paradigms such as: document-oriented databases (e.g., MongoDB) storing each child’s complete record as a structured document, graph databases emphasizing relationships between entities, or flat file structures storing data in CSV or similar formats.However, relational databases typically are used due to their maturity, queryoptimization capabilities, and broad tool support.

[0189] The normative reference database 400 employs indexing on frequently queried fields to optimize query performance. Specifically, in some configuration, indexes are created on: (i) age bin field, enabling rapid selection of all children in a specified age range; (ii) gender field, enablingDocket No. 1133.124WO1rapid filtering by male or female; (iii) test identifier field, enabling rapid selection of children who completed a specific test; and (iv) valid test flag, enabling rapid exclusion of invalid data. Database indexing trades increased storage space (as indexes require additional disk space) for dramatically improved query speed. Indexing facilitates sub-second retrieval of age-gender matched normative values even from databases containing tens of thousands of test sessions.

[0190] The normative reference database 400 also includes security measures to protect the confidentiality of children's information in compliance with privacy regulations such as the Health Insurance Portability and Accountability Act (HIPAA) in the United States or General Data Protection Regulation (GDPR) in Europe. Encryption may be applied at rest, where data stored on disk is encrypted using algorithms such as AES-256, and in transit, where data transmitted over networks is encrypted using protocols such as TLS. Access controls restrict database access to authorized personnel only, and audit logs track all access to provide security monitoring. For databases used in research contexts, additional protections such as de-identification (removal or coding of directly identifying information) and institutional review board oversight ensure ethical data handling.

[0191] Examples of the normative reference database 400 can be deployed in different infrastructure configurations depending on the use case and institutional preferences. In a local deployment, the normative reference database 400 resides on storage hardware at the testing site, such as on the storage 150 of computing system 110 shown in FIG. 1. Also in a local deployment, the normative reference database 400 provides data access without requiring network connectivity and ensuring that sensitive data remains within the institution's physical control. In a cloud deployment, the nonnative reference database 400 resides on remote servers accessed via network connection, such as through the network interface 155 shown in FIG, 1, This provides centralized maintenance of a single authoritative normative database accessed by multiple testing sites, automatic updatesDocket No. 1133.124WO1distributed to all sites, and reduced local storage requirements. Hybrid deployments may cache normative values, which are frequently accessed, locally while maintaining the master database in the cloud. The choice of deployment model balances considerations of data security, network reliability, update management, and regulatory requirements.Normative Reference Database Operation

[0192] The operation of the normative reference database 400 supports the social attention assessment system 100 as follows. During system initialization or periodically (e.g., monthly), the social attention assessment system 100 updates the normative reference database 400 by incorporating new validated data from children who have undergone testing. For each new child meeting inclusion criteria, their demographics 470, test metrics 475, and quality flags 480 are added to the normative reference database 400. The statistical reference values 440 are recalculated for affected age-gender strata, incorporating the new data points into means, standard deviations, and percentile distributions. The social attention assessment system 100 increments database version numbers and logs changes maintain an audit trail.

[0193] When an individual child undergoes assessment, the classification engine 175 or report generator 178 initiates a database query to retrieve age-gender matched normative values. The query specifies the child's age bin and gender. The normative reference database 400 returns the full set of statistical reference values 440 forthat age-gender stratum, covering all metrics of all tests.

[0194] For each metric that the assessed child generated, the age-gender matched comparison 494 executes to retrieve the child's metric value, retrieve the corresponding normative mean and SD, calculate the child’s z-score, determine percentile rank by comparing the child's value to the stored percentile distribution or by converting the z-score to percentile using standard normal distribution tables (for approximately normal distributions) or by empirical lookup (for non-normal distributions), and outputs the percentile rank, typically rounded to the nearest integer percentile (0-100).Docket No. 1133.124WO1This comparison process is repeated for each metric of each test the child completed. For a child completing all six tests generating 9 metrics each, a total of 54 percentile ranks are calculated, each representing the child's standing on a specific social attention capability relative to age-gender matched peers. These percentile ranks are then aggregated in various ways. By way of example and not limitation, aggregation can be domain-level percentile summaries averaging across all visual tests, all auditory tests, or the shifting test, the identification of metrics falling below the 25th percentile as potential areas of concern, and the overall social attention profile characterizing strengths and weaknesses across domains.

[0195] The retrieved statistical reference values 440 and calculated percentile ranks are passed to the report generator 178 for inclusion in the clinical assessment report 194. This provides quantitative, normatively referenced information to clinicians and caregivers about the child's social attention development.Alternative Embodiments and Optional Features

[0196] Several alternative embodiments and variations of the normative reference database 400 can be implemented while maintaining the core principle of age-gender stratified normative comparison. In one alternative example, age stratification employs narrower or wider age bins than the 4- month intervals described. For example, 2-month age bins provide finer developmental resolution, particularly valuable in the 12-24 month range when development is most rapid, though this increases the number of strata and may reduce sample sizes per stratum. Conversely, 6-month age bins reduce the number of strata and increase samples per stratum but may group together children at different developmental stages. In another example, age bins are defined adaptively based on empirical analysis of developmental trajectories, using wider bins during periods of slower development (e.g., 36-48 months) and narrower bins during periods of rapid change (e.g., 12-24 months).Docket No. 1133.124WO1|0197| In another alternative implementation, additional stratification factors beyond age and gender are employed, such as: stratification by language exposure (monolingual vs. bilingual children), by ethnicity or race (recognizing potential cultural differences in social attention norms), by geographic region, by socioeconomic status, or by other factors that might systematically affect social attention patterns. Each additional stratification factor, however, exponentially increases the number of strata. For example, adding 3 levels of ethnicity stratification to 9 age bins time 2 genders yields 54 strata instead of 18 when using only 2 levels. This can rapidly reduce sample sizes per stratum below levels adequate for stable statistical estimation. Therefore, examples of the social attention assessment system 100 typically use age and gender stratification alone, though additional stratification may be implemented when sample sizes permit.

[0198] The specific inclusion / exclusion criteria defining the neurotypical reference population may be adjusted to create alternative normative reference databases. For example, a more restrictive definition might require additional screening (e.g., genetic testing to exclude children with high-risk genetic variants, more comprehensive developmental testing) to ensure an exceptionally homogeneous typical population, though this increases assessment burden and cost. A more inclusive definition might include children with mild delays who do not meet criteria for developmental disorder diagnoses, creating norms that reflect community populations including subtle variability, though this may reduce sensitivity for detecting atypical patterns.

[0199] In some examples, the statistical methods used for percentile calculation can vary. The described approach uses empirical percentiles calculated directly from the observed distribution. Alternative approaches include parametric percentiles calculated by fitting a parametric distribution (e.g., normal, log-normal) to the data and using the fitted distribution's percentiles, quantile regression approaches that model percentile values as functions of age rather than binning by age, or Bayesian approaches that incorporate prior information about developmental trajectories. Typically, examples of the social attention assessment system 100 use empiricalDocket No. 1133.124WO1percentiles because they make no distributional assumptions and directly reflect observed data. Parametric approaches, however, may provide smoother percentile estimates when sample sizes are modest.Overall Social Attention Assessment Method

[0200] FIG. 5 illustrates a detailed flowchart of the overall method for assessing social attention in young children using eye-tracking technology. FIG. 5 depicts the procedural flow from initial system preparation through test administration, data collection, analysis, and clinical reporting, The social attention assessment method provides an objective, standardized approach for quantifying social attention capabilities across multiple domains and generating percentile-based characterizations to provide information about where a child's performance falls relative to age-matched and gender-matched normative reference populations. The social attention assessment method is designed for administration to toddlers aged 12 to 48 months in clinical settings such as pediatric offices, developmental clinics, preschool screenings, or research laboratories. Moreover, the social attention assessment method can be completed in as little as 10 to 20 minutes, including setup, calibration, and testing.

[0201] FIG. 5 depicts the method of the social attention assessment system 100 as a sequence of interconnected steps organized into several functional phases. These phases include system preparation and calibration ensuring technical readiness for valid data collection, test battery’ administration wherein the child views stimuli while gaze is tracked, data processing wherein raw gaze coordinates are transformed into quantitative social attention metrics, normative comparison wherein the child's metrics are compared to age-gender matched reference data, classification wherein dual analytical approaches generate social attention scores, positive predictive value determination based on patterns of test failure, and report generation wherein results are compiled and communicated to clinicians and caregivers. The social attention assessment method includes decision points that implement quality control (calibration acceptance) and diagnostic logic (evaluation of test failure patterns), as well as an optional iterative loopDocket No. 1133.124WO1enabling repeated assessment for longitudinal monitoring or test-retest reliability evaluation.System Preparation and Calibration

[0202] The method begins by setting up the social attention assessment system 100 and performing patent registration (operation 500). Operation 500 represents the initiation of an assessment session for a child.Commencement of the method may be triggered by a clinician or technician launching assessment software application on the computing system 110 and selecting a function such as " New Assessment" or " Begin Testing." System setup includes powering on hardware components (such as the eye-tracking device 125, display device 130, and audio output 135 as shown in FIG. 1), launching the assessment software on computing system 110, verifying that all hardware is functioning correctly (such as eye-tracker responsive, display showing correct output, audio playing at correct volume), and ensuring appropriate environmental conditions (adequate lighting for eye -tracking, minimal background noise, comfortable temperature). Patient registration includes entering the child's demographic information such as name or identifier, date of birth (from which age in months is calculated), gender, and other relevant information. This demographic data is used both for immediate age-gender matching during the current assessment and is stored in the patient data store 190 for future reference.

[0203] The social attention assessment method then proceeds to calibrate the eye-tracking device 125 for the patient (operation 505). This preparation establishes the mathematical relationship between measured eye features (pupil position, corneal reflection position) and point-of-gaze coordinates on the display screen 305, Calibration is used to ensure accurate gaze tracking because eye anatomy varies across individuals (interpupillary distance, corneal curvature, eye position relative to tracking cameras), and these individual differences must be characterized before gaze coordinates can be accurately estimated.Docket No. 1133.124WO1

[0204] In some examples, the calibration procedure implements a 5-point calibration protocol. Also known as a 5 -target calibration or multi-point calibration, this calibration procedure displays visual targets (such as animated cartoon characters, moving dots, or flashing shapes accompanied by sounds) sequentially at five locations on the display screen 305: center, upper-left, upper-right, lower-left, and lower-right. The child's attention is drawn to each target through animation and sound, and while the child fixates on each target, the eye-tracking device 125 captures images of the child's eyes and records the pupil and corneal reflection positions associated with that known screen location. After each of the five targets has been fixated, the social attention assessment system 100 calculates calibration parameters (mathematical transformation mapping eye features to screen coordinates). The social attention assessment system 100 also evaluates calibration quality by comparing estimated gaze locations to known target locations, calculating accuracy (mean angular error) and precision (variability of repeated estimates).

[0205] The social attention assessment system 100 assess whether the calibration is suitable. This is achieved by evaluating whether the achieved calibration quality meets acceptance criteria. The calibration quality assessment determines whether calibration is acceptable based on accuracy and precision metrics. In some examples, acceptable calibration requires accuracy better than 2 degrees of visual angle, while in other examples the required accuracy is better than 1.5 degrees, while in alternative examples the required accuracy is better than 1 degree. This is calculated as the mean angular deviation between estimated gaze points and actual target locations across the 5 calibration points. Precision is evaluated by requiring that repeated gaze estimates at each target (when available) show variability less than between 0.5 to 1 degree. These acceptance criteria are selected because: (i) accuracy of 2 degrees or better ensures that gaze estimates are sufficiently precise to determine which area of interest (typically 100-300 pixels width) the child is fixating; (ii) worse accuracy (e.g., >3 degrees) may result in misclassification of fixation locations particularly near AOI boundaries; and (iii) toddler eye-tracking is inherently noisier than adult eye-tracking due to greater head movement and less consistent fixation behavior, so overlyDocket No. 1133.124WO1stringent criteria (e.g., <0.5 degrees) would result in excessive calibration failures.

[0206] Once the calibration quality is acceptable, then the method proceeds. But if the calibration quality is not acceptable, a recalibration loop revises and repeats the calibration procedure. This recalibration loop may iterate multiple times until acceptable calibration is achieved. If acceptable calibration cannot be achieved after multiple attempts, the session may be terminated and rescheduled. This outcome, however, is relatively rare (occurring in less than 10% of sessions, or more specifically less than 5%) with proper technique and in appropriately selected children.Test Battery Administration and Data Collection

[0207] Following successful calibration, the social attention assessment method proceeds to administer the test battery (operation 510). This is where the child views eye-tracking test stimuli while gaze data is captured.Operation 510 implements the test battery 165 functionality in coordination with the eye-tracking device 125, display device 130, and audio output 135. Administration of the test battery 165 includes presenting at least one test from the six-test battery described in FIG. 2 and above, or more typically at least two tests from different domains, or more specifically at least three tests, or even more specifically all six tests described above. Tests may be presented in a predetermined sequence (e.g., visual tests first, then auditory, then shifting), in randomized order, or in an adaptive sequence based on initial results. Each test runs for its specified duration (58 to 95 seconds depending on the specific test), during which visual stimuli are displayed on the display device 130, audio stimuli are delivered through speakers (for auditory and joint attention tests), and the child’s gaze is continuously tracked.

[0208] The child is not given explicit instructions to look at specific stimuli or to perform any task. Rather, the child views the stimuli naturally according to their spontaneous preferences and attention patterns. This naturalistic, task-free approach is appropriate for toddlers who may notDocket No. 1133.124WO1reliably follow explicit instructions and ensures that measured gaze patterns reflect the child’s intrinsic social attention biases rather than compliance with instructions.

[0209] Between tests, brief inter-test intervals (typically between about 5 to 15 seconds) provide transition time and allow the child to reset attention. During these inter-test intervals, a central fixation stimulus (such as a flashing star or animated character) may be displayed to re-center the child's gaze before the next test begins.

[0210] The social attention assessment method then proceeds to collect raw gaze data using the eye-tracking device 125 (operation 515). This occurs concurrently with operation 510 but is shown as a separate step in FIG. 5 to emphasize the data acquisition function. During each test presentation, the eye-tracking device 125 continuously captures gaze coordinates at the system's sampling rate (typically 60 to 600 Hz and in some examples 120 Hz), generating a time series of x-y screen coordinates with timestamps indicating where the child is looking at each sampling instant. This raw gaze data stream is transmitted to the computing system 110 and temporarily buffered in memory 145 during real-time collection, and then written to storage 150 following test completion. Raw data files for each test typically range from 100 KB to 5 MB depending on sampling rate, test duration, and data format.

[0211] Quality monitoring occurs concurrently with operation 515, wherein the social attention assessment system 100 tracks indicators such as tracking loss periods (times when gaze could not be detected), total looking time (time when child is looking at screen vs. away), and calibration drift (gradual degradation of calibration accuracy during testing due to head movement). Excessive tracking loss (e.g., >30% of test duration) or insufficient looking time (e.g., <40% of test duration) may trigger real-time alerts to the technician or may result in flagging the test for potential exclusion during post -processing.Docket No. 1133.124WO1Feature Extraction and Metric Calculation

[0212] Following data collection in operation 515, the social attention assessment method proceeds to extract features and metrics (operation 520). Operation 520 transforms raw gaze coordinate time series into quantitative social attention metrics. Feature extraction includes several sub-processes operating on the raw gaze data. First, data filtering removes artifacts and invalid data points, including blink periods (identified as intervals when pupils cannot be detected), data points with low tracking confidence scores (provided by some eye-tracking systems), gaze coordinates falling outside the screen boundaries (indicating off-screen looks); and fixations with durations less than a minimum threshold (typically less than 100 ms), which likely represent noise or saccade periods rather than true fixations. This filtering reduces noise in the gaze data stream while preserving genuine fixation events.

[0213] Second, an area of interest (AOI) mapping assigns each gaze coordinate to a defined spatial region representing different stimulus types. For preferential looking tests (GeoPref 215, Complex Social 220, Outside Play 225, Motherese vs. Highway 2352, Motherese vs. Techno 240), AOIs are defined as social stimulus region (left or right side of screen), non-social stimulus region (opposite side), and neutral or off-screen regions. For the Joint Attention test 250, AOIs are defined as target objects (items indicated by the actress), non-target objects, actress's face, and background. Each gaze coordinate is classified as falling within one of these AOIs based on spatial comparison of the coordinate to AOI boundaries.

[0214] Third, fixation event identification occurs, which groups consecutive gaze coordinates that remain within a small spatial threshold (typically 2 degrees of visual angle, or approximately 30-50 pixels) into discrete fixation events, calculating duration, location, and onset / offset times for each fixation. Saccades (rapid eye movements between fixations) are identified as intervals between fixations characterized by high velocity gaze changes. Fourth, a metric calculation computes the 9 standardized metrics for each test by aggregating and statistically summarizing the fixation andDocket No. 1133.124WO1saccade data. For example, percent fixation on social stimuli is calculated as: (sum of all fixation durations within social AOIs) / (total looking time) x 100. Number of saccades per second within social AOIs is calculated as: (count of saccades with onset in social AOI) / (total duration spent in social AOI). These and other metrics are calculated following standardized formulas ensuring consistency across test sessions and enabling valid comparison to normative data.

[0215] The output of operation 520 includes a feature vector, also referred to as a metric vector or data vector, containing the calculated metric values for each test the child completed. For a child completing all six tests, this feature vector contains 54 values (9 metrics × 6 tests), or fewer values for partial battery completion. This feature vector is passed to subsequent classification operations for analysis.Normative Data Retrieval and Age-Gender Matching

[0216] The social attention assessment method proceeds to retrieve age-gender matched nonns (operation 525). Operation 525 implements the database query and normative comparison functionality that enables percentile ranking. Operation 525 accesses the normative reference database 400 to retrieve statistical reference values for the appropriate comparison group. Specifically, operation 525 does the following: (i) determines the child's age in months from date of birth and test date; (ii) identifies the corresponding age bin based on the stratification structure 420, such as determining that a child aged 22 months falls into the 20-23 month age bin; (iii) retrieves the child's gender from the registration data entered in operation 500; (iv) queries the database for statistical reference values 440 (mean, standard deviation, and percentile distributions) corresponding to that age bin and gender; and (v) retrieves reference values for all metrics of all tests, ensuring that subsequent comparisons use age-appropriate and gender-appropriate norms. Operation 525 executes rapidly, typically within 100 milliseconds to 1 second, due to database indexing that optimizes query performance. The retrieved normative values are loaded into memory 145 where they are accessible to the classification engine 175 for subsequent comparison operations.Docket No. 1133.124WO1

[0217] In some examples, operation 525 also retrieves additional contextual information from the normative reference database 400 such as sample sizes for the matched age-gender group (enabling assessment of statistical confidence), distributional characteristics (skewness, kurtosis), or even raw data points from the matched group (enabling advanced visualization showing the child's position within the scatter of normative data).Dual Classification Analysis

[0218] The social attention assessment method proceeds to select a classification type (operation 530). Examples of the social attention assessment system 100 and method include two distinct analytical approaches that can be applied in either one or the other or together parallel to generate complementary social attention assessments. This dual approach classification provides both transparency (through the simple-threshold method) and sophisticated multi-feature analysis (through the machine learning method), addressing different clinical and research needs.

[0219] Depending on the classification type selected, operation 530 branches into a simple -threshold classifier (operation 535) path, a machine learning classifier (operation 540) path, or into two parallel processing paths that use both classifiers. Operation 535, the simple-threshold classifier, uses a straightforward, transparent classification approach based on comparing individual metrics to empirically-derived thresholds. The simple-threshold classifier (operation 535) operates by: (i) for each test the child completed, identifying the primary’ threshold metric (percent fixation on non-social stimuli for visual and auditory tests, percent fixation on non -target objects for joint attention test); (ii) comparing each metric value to the predetermined threshold (69% for visual / auditory tests, 30% for joint attention test); (iii) classifying each test as "failed" if the me tric exceeds the threshold or "passed" if it does not; (i v) identifying the child's worst performance (highest percent non-social or non-target value) across all tests; (v) calculating a preliminary’ score equal to this worst percentage value; (vi) adding 5 points to the score for each additional test that was also failed,Docket No. 1133.124WO1implementing a multiple-test-failure penalty that reflects increased severity when multiple domains show atypical patterns; (vii) capping the maximum score at 100; and (viii) if the calculated score is 69 or higher (indicating at least one test failed), applying a linear transformation to map the score to a 50-100 scale, wherein 50 represents the threshold level (69% non-social fixation) and 100 represents the maximum possible atypicality. This transformation generates a simple-threshold score ranging from 0 to 100, where 0-50 indicates typical social attention and 51-100 indicates atypical social attention with increasing scores reflecting increasing atypicality.

[0220] The simple-threshold classifier (operation 535) is advantageous because it is transparent and interpretable. Clinicians can understand exactly how the score was derived. Moreover, it relies on well-validated thresholds that achieve high specificity (>95%) based on analysis of over 2,000 children. It also emphasizes the worst performance, ensuring that children failing even one test are detected. And it accounts for multiple test failure through the 5-point penalty, capturing the clinically meaningful patern that children failing multiple tests show more severe atypicality.

[0221] The machine learning classifier (operation 540) implements a machine learning classification approach that simultaneously considers multiple metrics across multiple tests to generate a probabilistic classification score. The machine learning classifier (operation 540) includes a trained statistical model, such as a logistic regression model, random forest model, support vector machine, gradient boosting model, or other supervised learning algorithm, that has been trained on a dataset of children with known diagnostic outcomes (autism spectrum disorder vs. no autism) to learn the multivariate patterns of eye-tracking metrics that distinguish typical from atypical social attention. The machine learning classifier (operation 540): (i) assembling the child's complete feature vector containing all metrics from all completed tests; (ii) applying feature normalization or standardization to ensure metrics on different scales are appropriately weighted (typically using robust scaling that divides each metric by its interquartile range, or z-score normalization dividing by standard deviation); (iii) inputting the normalizedDocket No. 1133.124WO1feature vector to the trained ML model; (iv) receiving the model's probability output, representing the model's estimated probability that the child has atypical social attention or autism based on the input features; and (v) transforming the probability (ranging 0.0 to 1.0) to a score scale of 0 to 100 by multiplying by 100.

[0222] In some examples, the machine learning classifier (operation 540) includes a logistic regression model that has been trained via 5 -fold cross-validation on data from over 2,400 children, optimized to maximize area under the receiver operating characteristic curve (AUC) while maintaining specificity of at least 90%, or more specifically at least 92%. The logistic regression model learns optimal weights for each metric, automatically discovering which features are most predictive and how features interact. Alternative examples can use random forest models, which aggregate predictions from multiple decision trees. Other examples can use gradient boosting models, which sequentially fit trees to residual errors; or neural network models. The choice of algorithm reflects a trade-off between performance (random forests and gradient boosting sometimes achieve slightly higher AUC), interpretability (logistic regression coefficients are interpretable as feature importance weights), and computational efficiency (logistic regression is faster for prediction).

[0223] The technical effect of the dual classification approach is that clinicians receive both a transparent, explainable score (simple-threshold) and a sophisticated, optimized score (machine learning), and can consider agreement or disagreement between the two approaches when interpreting results. In some implementations, high agreement (both classifiers giving similar scores) increases confidence in classification, while disagreement (classifiers giving substantially different scores, such as differing by more than 20 points) may prompt additional evaluation or re-testing.Consensus Determination and Test Failure Analysis

[0224] The next operation is to determine consensus (operation 545), wherein the outputs from the simple-threshold classifier (operation 535) and the machine learning classifier (operation 540) are integrated. Operation 545Docket No. 1133.124WO1is shown as an optional by the dashed lines, and is only used when both of the simple-threshold classifier (operation 535) and the machine learning classifier (operation 540) are used. Assuming a configuration that uses both of the classifiers, and depending on the implementation, the consensus determination (operation 545) can include one or more of the following: averaging the two scores to generate a combined score; selecting the higher score (indicating worse social attention) as a conservative approach; selecting the lower score as a liberal approach; or reporting both scores separately without combining them. In one example, both scores are reported separately in the assessment report 194, enabling clinicians to see both perspectives. In addition, a consensus score can be calculated (for example, as the average of the two) for use in decision-making or research applications.

[0225] Following the consensus determination (operation 545), the social attention assessment method proceeds to count the number of failed tests (operation 550). This is a main analysis operation that evaluates the pattern of test failures across the multi-test battery. Operation 550 determines how many of the tests that the child has completed resulted in failure classification. This failure is based on the threshold criteria applied in the simple-threshold classifier (operation 535). Specifically, for each test, if percent fixation on non-social stimuli exceeded 69% (for visual / auditory tests) or if percent fixation on non-target exceeded 30% (for joint attention), that test is counted as failed. The total number of failed tests is tallied in operation 550, yielding a count ranging from 0 (no tests failed, indicating typical social attention across all assessed domains) to 6 (all tests failed, indicating severe social attention impairment across all domains).

[0226] The count of failed tests (operation 550) is useful because validation data demonstrates that the number of failures strongly predicts diagnostic outcome. Specifically, an empirical analysis of 448 children who completed 5 or 6 tests revealed the following. Children failing exactly I test had 85% positive predictive value (PPV) for autism diagnosis, children failing exactly 2 tests had 98% PPV, and children failing 3 or more tests had 100% PPVDocket No. 1133.124WO1(zero false positives in the validation sample). This dose-response relationship, wherein increasing test failures progressively increase diagnostic confidence, provides valuable clinical information beyond what is available from any single test or from classification scores alone.Positive Predictive Value Determination

[0227] The social attention assessment method then reaches a decision point as to whether the number of failed test is greater than 3 (operation 555). Operation 555 evaluates whether the child failed three or more tests from the test battery. Operation 555 implements logic for flagging cases with exceptionally high diagnostic confidence based on the multiple-test- failure pattern. If the child failed 3 or more tests, meaning that operation 555 returns a “YES”, then the social attention assessment method concludes that there is a high social attention disorder (SAD) likelihood (operation 560). In other words, this particular child is flagged as having particularly high probability of social attention disorder diagnosis. The high SAD likelihood of operation 560 is based on empirical validation data showing 100% positive predictive value for children failing >3 tests, as described above. This flag is incorporated into the assessment report 194, typically with prominent visual indication and accompanying text such as, by way of example and not limitation: " Multiple test failures (3 or more tests) are strongly associated with social attention disorder (including autism spectrum disorder). Positive predictive value for SAD diagnosis: 100% based on validation sample. Immediate referral for comprehensive diagnostic evaluation is recommended."

[0228] The 100% PPV finding for >3 test failures represents an unexpected result that provides exceptional clinical value. It was not a priori obvious that multiple test failures would yield perfect predictive value. One might have expected that some neurotypically developing children or children with non-autism delays might occasionally fail 3 or more tests due to chance, inattention, or other factors. The empirical finding that zero such false positives occurred in a validation sample of 448 children indicates that the pattern of failing >3 tests from this multi-domain battery is highly specificDocket No. 1133.124WO1for autism or severe social attention impairment, providing a "rule-in" criterion wherein presence of this pattern virtually confirms the diagnosis pending formal clinical evaluation.

[0229] If the child failed fewer than 3 tests, meaning that operation 555 returns a “NO”, then the social attention assessment method bypasses the high SAD likelihood flag (operation 560). This includes children failing 0, 1, or 2 tests, as well as children who completed fewer than 3 tests total (making >3 failures impossible). These children proceed directly to report generation (operation 565) without the special high-likelihood flagging. It should be noted, however, that their classification scores and test-specific results are still reported and may indicate elevated likelihood depending on the specific scores.Report Generation and Data Archiving

[0230] Operation 565implements the functionality of the report generator 178, which compiles the complete assessment results into a formatted clinical document. Report generation includes one or more of the following operations. First, a percentile rank calculation transforms each of the child’s metric values into percentile scores by comparing to the age-gender matched normative distributions retrieved in operation 525. For each metric, the social attention assessment system 100 determines what percentage of the normative sample scored at or below the child's value, expressing this as a percentile from 0 to 100. For example, if the child's percent fixation on social images is lower than 85% of age-gender matched children, the percentile rank is 15 (15th percentile). This calculation is repeated for all metrics from all tests.

[0231] Second, domain-specific summaries aggregate metrics within each of the three domains (visual social attention, auditory social attention, shifting social attention) to generate domain-level percentile scores. Domain summarization may be implemented by averaging percentile ranks across all tests within a domain, by averaging raw metric values within a domain before percentile ranking, or by other aggregation methods.Docket No. 1133.124WO1

[0232] Third, classification score reporting includes the simple-threshold score from the simple-threshold classifier (operation 535) and the machine learning score from the machine learning classifier (operation 540), both expressed on the 0-100 scale. If applicable, the high SAD likelihood flag from operation 560 is prominently indicated.

[0233] Fourth, the social attention assessment system 100 can generate graphical visualizations illustrating the child's percentile ranks across metrics. Often these illustrations use formats such as horizontal bars showing percentile rank from 0-100 with marked regions for below-average (<25th percentile), average (25-75th percentile), and above-average (>75th percentile), line graphs connecting percentile ranks across multiple metrics to show the profile shape, or scatter plots showing the child's position (marked point) overlaid on the distribution of normative data points.

[0234] Fifth, the social attention assessment system 100 can generate clinical interpretation text based on rule-based logic. This rule-based logic can include flagging percentile ranks below 25 as "below average" or "area of concern,’’ suggesting referrals for further evaluation when multiple domains show below-average performance, recommending specific interventions (e.g., speech therapy if auditory social attention is low, social skills groups if visual / shifting attention is low).

[0235] Some examples of the social attention assessment method output a complete compiled report. This report is typically formatted as a multi -page PDF document or as an HTML page, and can include a header with child identification and test date, individual test results section showing metric values and percentiles for each test, domain summary section showing aggregate domain scores, classification scores and likelihood indicators, graphical visualizations, and clinical interpretation and recommendations. The report is structured for review by pediatricians, psychologists, speechlanguage pathologists, educators, or parents, using clinical terminology and visual presentations that communicate findings clearly to non-expert audiences.Docket No. 1133.124WO1

[0236] In some examples, the social attention assessment method archives the complete assessment data in the patient data store 190 for future reference, longitudinal tracking, and potential research applications. By way of example, stored data can include one or more of raw gaze data files, extracted metric values, retrieved normative reference values, classification scores, generated report document, quality flags, and session metadata (date, time, technician, testing location). Data storage enables retrieval of prior assessment results when the child returns for follow-up, supports test-retest reliability analysis by linking multiple sessions for the same child, and provides data for ongoing research to validate the assessment method in new populations or to develop enhanced analytical approaches.Method Completion and Optional Iteration

[0237] Following report generation and data storage, the social attention assessment method in some examples is completed and the assessment session is over. The clinician or technician reviews the generated report, discusses results with the child's caregivers, and makes referrals or recommendations as appropriate based on the findings. In alternative examples, an optional retest loop (or optional iteration) returns the method to operation 500 for repeated assessment of the same or a different child. This retest loop may be used in several different scenarios, including immediate re-testing if initial data quality was poor, scheduled follow-up assessment after a time interval (e.g., 6 months later) to monitor developmental progress, post-intervention assessment to evaluate treatment effects, or test- retest reliability assessment wherein the same child is tested multiple times to evaluate measurement stability. The social attention assessment method supports unlimited iteration through this retest loop. This facilitates longitudinal assessment of individual children or sequential assessment of multiple children in a clinic or research setting.Alternative Embodiments and Method Variations

[0238] Several alternative embodiments and variations of the social attention assessment method may be implemented while maintaining the core procedural framework. In one alternative example, abbreviated testingDocket No. 1133.124WO1is implemented wherein fewer than six tests are administered, such as only 2, 3, or 4 tests selected from the battery. Test selection may be predetermined (e.g., always administering GeoPref 215, Motherese vs. Highway 235, and Joint Attention 250 for three-test rapid screening), may be adaptive based on initial results (e.g., administering tests from domains showing borderline performance on initial tests to gain more information in those domains), or may be based on available time or child cooperation. Abbreviated testing reduces total assessment duration (potentially to 5-10 minutes for 2-3 tests) but also reduces sensitivity for detecting atypical social attention, as the multiple-test-failure criterion becomes less powerful with fewer tests.Nonetheless, abbreviated batteries maintain high specificity and may be appropriate for screening applications where time is limited.

[0239] In another alternative implementation, the calibration procedure of operation 505 uses alternative calibration protocols. These alternative calibration protocols include 3 -point calibration, which is faster but less accurate, 9-point calibration (more comprehensive but more time-consuming), or infant-adaptive calibration wherein target locations and presentations are optimized specifically for infant attention characteristics. In a typical implementation of the social attention assessment method the 5-point calibration is used as it balances accuracy and efficiency for the toddler population.

[0240] The order of operations may be varied in some configurations. For example, the normative data retrieval in operation 525 could occur earlier in the method flow (such as immediately after the patient registration in operation 500) rather than after feature extraction (operation 520), thereby pre-loading reference data before testing begins. Feature extraction (operation 520) could occur in real-time during test administration (operation 510) and data collection (operation 515) rather than as a postprocessing operation. This facilitates a live display of metrics as they are calculated. It should be noted, however, that real-time extraction increases computational demands and may interfere with stimulus presentation if processing resources are limited.Docket No. 1133.124WO1

[0241] As noted above, the selection of the classification type (operation 530) could be replaced with a single classification approach in simplified embodiments, using only simple-threshold classifier (operation 535) or only machine learning classifier (operation 540) rather than both. In some implementations, however, the dual approach is useful because it provides complementary information and the additional computational cost is negligible,

[0242] The >3 test failures criterion in operation 555 could be adjusted to different thresholds, such as >2 test failures or >4 test failures, depending on the desired balance between sensitivity and specificity. The >3 criterion has been found to be useful based on empirical optimization, which shows that this threshold achieves 100% PPV while capturing approximately 30-40% of children with autism (those with more severe social attention impairments affecting multiple domains).

[0243] Report formatting and content can be customized based on intended audience or use case. Reports for clinicians may include more technical detail and statistical information, while reports for parents may emphasize plain-language interpretation and recommendations. Reports for research applications may include additional data such as raw metric values, z-scores, or detailed visualizations that are not necessary for clinical decision-making.Advantages of the Social Attention Assessment System and Method

[0244] Examples of the social attention assessment system 100 and method provide objective, quantitative measurement of social attention capabilities through a multi-domain test battery’ structure that captures heterogeneous developmental patterns across visual, auditory, and attention-shifting domains. The large, age-gender stratified normative database containing data from over 2,000 children enables precise percentile ranking that accounts for rapid developmental changes during toddlerhood and systematic sex differences in social attention development. The comprehensive quality control and multi-site collection ensure statistical stability and generalizability across diverse populations, while the efficient queryDocket No. 1133.124WO1mechanisms enable real-time percentile calculation supporting clinical workflow.

[0245] The rapid 10-20 minute administration by trained technicians dramatically reduces assessment burden while maintaining engagement through naturalistic, child-friendly video stimuli. The standardized metric extraction across multiple behavioral dimensions, including fixation patterns, saccadic movements, attention shifting cycles, and temporal dynamics, enables detailed characterization providing clinically useful information about specific strengths and weaknesses. The scalable design enables deployment across diverse settings including pediatric offices, preschools, and research laboratories without requiring specialized clinical expertise, addressing critical shortages in developmental assessment capacity.

[0246] The gaze-contingent auditory assessment represents a unique capability that measures active, sustained preference for social speech through interactive control, engaging children's attention while providing objective measurement of auditory social attention in very young children (12-24 months) who cannot perform tasks requiring explicit verbal responses. The percentile-based reporting generates clinically interpretable continuous scores familiar to healthcare providers, supporting multiple applications including preschool readiness assessment, intervention planning targeting specific domains (auditory versus visual social attention), progress monitoring through repeated assessments, and early identification of children who may benefit from speech therapy or social skills training at ages (12-48 months) when intervention is most effective due to brain plasticity.Computation Environment

[0247] FIG, 6 illustrates a diagrammatic representation of a computing system within which instructions may be executed for causing a machine to perform any one or more of the methodologies discussed herein. By way of example and not limitation, the following components discussed in thisDocket No. 1133.124WO1disclosure can be considered a computing system, including the hardware components 105, the eye-tracking device 125, the processor 140, and the computing system 110, just to name a few.

[0248] Specifically, FIG. 6 shows a diagrammatic representation of the machine 600 in the example form of a computing system, within which instructions 608 (e.g., software, a program, an application, an applet, an app, or other executable code) for causing the machine 600 to perform any one or more of the methodologies discussed herein may be executed. The instructions 608 transform the general, non-programmed machine 600 into a particular machine 600 programmed to carry out the described and illustrated functions in the manner described. In alternative examples, the machine 600 operates as a standalone device or may be coupled (e.g., networked) to other machines. In a networked deployment, the machine 600 may operate in the capacity of a server machine or a client machine in a server-client network environment, or as a peer machine in a peer-to-peer (or distributed) network environment. The machine 600 may comprise, but not be limited to, a server computer, a client computer, a personal computer (PC), a tablet computer, a laptop computer, a netbook, a set-top box (STB), a PDA, an entertainment media system, a cellular telephone, a smart phone, a mobile device, a wearable device (e.g., a smart watch), a smart home device (e.g., a smart appliance), other smart devices, a web appliance, a network router, a network switch, a network bridge, or any machine capable of executing the instructions 608, sequentially or otherwise, that specify actions to be taken by the machine 600. Further, while only a single machine 600 is illustrated, the term “machine” shall also be taken to include a collection of machines 600 that individually or jointly execute the instructions 608 to perform any one or more of the methodologies discussed herein.

[0249] The machine 600 may include processors 602, memory 604, and I / O components 642, which may be configured to communicate with each other such as via a bus 644. In an example, the processors 602 (e.g., a Central Processing Unit (CPU), a Reduced Instruction Set Computing (RISC) processor, a Complex Instruction Set Computing (CISC) processor, aDocket No. 1133.124WO1Graphics Processing Unit (GPU), a Digital Signal Processor (DSP), an ASIC, a Radio-Frequency Integrated Circuit (RFIC), another processor, or any suitable combination thereof) may include, for example, a processor 606 and a processor 610 that may execute the instructions 608. The term “■processor” is intended to include multi-core processors that may comprise two or more independent processors (sometimes referred to as “cores”) that may execute instructions contemporaneously. Although FIG. 6 shows multiple processors 602, the machine 600 may include a single processor with a single core, a single processor with multiple cores (e.g., a multi-core processor), multiple processors with a single core, multiple processors with multiples cores, or any combination thereof.

[0250] The memory 604 may include a main memory 612, a static memory 614, and a storage unit 616, both accessible to the processors 602 such as via the bus 644. The main memory 604, the static memory 614, and storage unit 616 store the instructions 608 embodying any one or more of the methodologies or functions described herein. The instructions 608 may also reside, completely or partially, within the main memory 612, within the static memory 614, within machine-readable medium 618 within the storage unit 616, within at least one of the processors 602 (e.g., within the processor’s cache memory), or any suitable combination thereof, during execution thereof by the machine 600.

[0251] The I / O components 642 may include a wide variety of components to receive input, provide output, produce output, transmit information, exchange information, capture measurements, and so on. The specific I / O components 642 that are included in a particular machine will depend on the type of machine. For example, portable machines such as mobile phones will likely include a touch input device or other such input mechanisms, while a headless server machine will likely not include such a touch input device. It will be appreciated that the I / O components 642 may include many other components that are not shown in FIG. 6. The I / O components 642 are grouped according to functionality merely for simplifying the following discussion and the grouping is in no way limiting. In various examples, theDocket No. 1133.124WO1I / O components 642 may include output components 628 andinput components 630. The output components 628 may include visual components (e.g., a display such as a plasma display panel (PDP), a light emitting diode (LED) display, a liquid crystal display (LCD), a projector, or a cathode ray tube (CRT)), acoustic components (e.g., speakers), haptic components (e.g., a vibratory motor, resistance mechanisms), other signal generators, and so forth. The input components 630 may include alphanumeric input components (e.g., a keyboard, a touch screen configured to receive alphanumeric input, a photo-optical keyboard, or other alphanumeric input components), point-based input components (e.g., a mouse, a touchpad, a trackball, a joystick, a motion sensor, or another pointing instrument), tactile input components (e.g., a physical button, a touch screen that provides location and / or force of touches or touch gestures, or other tactile input components), audio input components (e.g., a microphone), and the like.

[0252] In further examples, the I / O components 642 may include biometric components 632, motion components 634, environmental components 636, or position components 638, among a wide array of other components. For example, the biometric components 632 may include components to detect expressions (e.g., hand expressions, facial expressions, vocal expressions, body gestures, or eye tracking), measure biosignals (e.g., blood pressure, heart rate, body temperature, perspiration, or brain waves), identify a person (e.g., voice identification, retinal identification, facial identification, fingerprint identification, or electroencephalogram-based identification), and the like. The motion components 634 may include acceleration sensor components (e.g., accelerometer), gravitation sensor components, rotation sensor components (e.g., gyroscope), and so forth. The environmental components 636 may include, for example, illumination sensor components (e.g., photometer), temperature sensor components (e.g., one or more thermometers that detect ambient temperature), humidity sensor components, pressure sensor components (e.g., barometer), acoustic sensor components (e.g., one or more microphones that detect background noise), proximity sensor components (e.g., infrared sensors that detect nearbyDocket No. 1133.124WO1objects), gas sensors (e.g., gas detection sensors to detection concentrations of hazardous gases for safety or to measure pollutants in the atmosphere), or other components that may provide indications, measurements, or signals corresponding to a surrounding physical environment. Theposition components 638 may include location sensor components (e.g., a GPS receiver component), altitude sensor components (e.g., altimeters or barometers that detect air pressure from which altitude may be derived), orientation sensor components (e.g., magnetometers), and the like.

[0253] Communication may be implemented using a wide variety of technologies. The I / O components 642 may includecommunication components 640 operable to couple the machine 600 to a network 620 or devices 622 via a coupling 624 and a coupling 626, respectively. For example, the communication components 640 may include a network interface component or another suitable device to interface with the network 620. In further examples, the communication components 640 may include wired communication components, wireless communication components, cellular communication components, Near Field Communication (NFC) components, Bluetooth® components (e.g., Bluetooth® Low Energy), Wi-Fi® components, and other communication components to provide communication via other modalities. The devices 622 may be another machine or any of a wide variety of peripheral devices (e.g., a peripheral device coupled via a USB).

[0254] Moreover, the communication components 640 may detect identifiers or include components operable to detect identifiers. For example, the communication components 640 may include Radio Frequency Identification (RFID) tag reader components, NFC smart tag detection components, optical reader components (e.g., an optical sensor to detect onedimensional bar codes such as Universal Product Code (UPC) bar code, multi-dimensional bar codes such as Quick Response (QR) code, Aztec code, Data Matrix, Dataglyph, MaxiCode, PDF417, Ultra Code, UCC RSS- 2D bar code, and other optical codes), or acoustic detection components (e.g., microphones to identify tagged audio signals). In addition, a variety ofDocket No. 1133.124WO1information may be derived via the communication components 640, such as location via Internet Protocol (IP) geolocation, location via Wi-Fi® signal triangulation, location via detecting an NFC beacon signal that may indicate a particular location, and so forth.Executable Instructions and Machine Storage

[0255] The various memories (i.e., memory 604, main memory 612, static memory 614, and / or memory of the processors 602) and / or storage unit 616 may store one or more sets of instructions and data structures (e.g., software) embodying or utilized by any one or more of the methodologies or functions described herein. These instructions (e.g., the instructions 608), when executed by processors 602, cause various operations to implement the disclosed examples.

[0256] As used herein, the terms “machine-storage medium,” “devicestorage medium,” “computer-storage medium” mean the same thing and may be used interchangeably in this disclosure. The terms refer to a single or multiple storage devices and / or media (e.g., a centralized or distributed database, and / or associated caches and servers) that store executable instructions and / or data in a non-transitory manner and that can be read by one or more processors. The terms shall accordingly be taken to include, but not be limited to, solid-state memories, and optical and magnetic media, including memory internal or external to processors. Specific examples of machine-storage media, computer-storage media and / or device-storage media include non-volatile memory, including by way of example semiconductor memory devices, e.g., erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), FPGA, and flash memory devices; magnetic disks such as internal hard disks and removable disks; magneto-optical disks; and CD- ROM and DVD-ROM disks. The terms “machine-storage media,” “computer-storage media,” and “device -storage media” specifically exclude carrier waves, modulated data signals, and other such media, at least some of which are covered under the term “signal medium” discussed below.Docket No. 1133.124WO1Transmission Medium

[0257] In various examples, one or more portions of the network 620 may be an ad hoc network, an intranet, an extranet, a VPN, a LAN, a WLAN, a WAN, a WWAN, a MAN, the Internet, a portion of the Internet, a portion of the PSTN, a plain old telephone service (POTS) network, a cellular telephone network, a wireless network, a Wi-Fi® network, another type of network, or a combination of two or more such networks. For example, the network 620 or a portion of the network 620 may include a wireless or cellular network, and the coupling 624 may be a Code Division Multiple Access (CDMA) connection, a Global System for Mobile communications (GSM) connection, or another type of cellular or wireless coupling. In this example, the coupling 624 may implement any of a variety of types of data transfer technology, such as Single Carrier Radio Transmission Technology (1xRTT), Evolution-Data Optimized (EVDO) technology. General Packet Radio Service (GPRS) technology, Enhanced Data rates for GSM Evolution (EDGE) technology, third Generation Partnership Project (3GPP) including 3G, fourth generation wireless (4G) networks, Universal Mobile Telecommunications System (UMTS), High Speed Packet Access (HSPA), Worldwide Interoperability for Microwave Access (WiMAX), Long Term Evolution (LTE) standard, others defined by various standard-setting organizations, other long range protocols, or other data transfer technology.

[0258] The instructions 608 may be transmitted or received over the network 620 using a transmission medium via a network interface device (e.g., a network interface component included in thecommunication components 640) and utilizing any one of a number of well-known transfer protocols (e.g., hypertext transfer protocol (HTTP)).Similarly, the instructions 608 may be transmitted or received using a transmission medium via the coupling 626 (e.g., a peer-to-peer coupling) to the devices 622. The terms “transmission medium” and “signal medium” mean the same thing and may be used interchangeably in this disclosure. The terms “transmission medium” and “signal medium” shall be taken to include any intangible medium that is capable of storing, encoding, or carrying the instructions 608 for execution by the machine 600, and includes digital orDocket No. 1133.124WO1analog communications signals or other intangible media to facilitate communication of such software. Hence, the terms ‘-transmission medium” and “signal medium” shall be taken to include any form of modulated data signal, carrier wave, and so forth. The term “modulated data signal” means a signal that has one or more of its characteristics set or changed in such a matter as to encode information in the signal.Additional Notes

[0259] The following, non-limited examples, detail certain aspects of the present subject matter to solve the challenges and provide the benefits discussed herein, among others.

[0260] Example l is a method for assessing social attention m a child, comprising: presenting to a child at least one eye-tracking test comprising visual or auditory stimuli; measuring, using an eye-tracking device, eye gaze metrics during presentation of the visual or auditory stimuli; comparing the measured eye gaze metrics to normative data from a plurality of children; and generating a percentile ranking for at least one of the measured eye gaze metrics, wherein the percentile ranking indicates where a social attention level of the child falls relative to the plurality of children,

[0261] In Example 2, the subject matter of Example 1 includes, wherein the at least one eye-tracking test comprises a multi-domain test battery including at least two tests selected from at least two different domains, the domains comprising: visual social attention tests presenting paired social visual stimuli and non-social visual stimuli; auditory social attention tests presenting social audio stimuli and non-social audio stimuli; and shifting social attention tests presenting joint attention probes,

[0262] In Example 3, the subject matter of Example 2 includes, wherein at least one auditory social attention test comprises gaze-contingent audio stimulus presentation, wherein infant-directed speech is activated when the child's gaze is directed to a first area of interest and non-social audio is activated when the child's gaze is directed to a second area of interest.Docket No. 1133.124WO1

[0263] In Example 4, the subject matter of Example 1 includes, wherein the normative data comprises eye gaze metrics from at least 2,000 children stratified by age bins of 4-month intervals and by gender, and wherein comparing the measured eye gaze metrics comprises selecting an age-matched and gender-matched comparison group from the normative data.

[0264] In Example 5, the subject matter of Example 1 includes applying a simple-threshold classifier that compares the measured eye gaze metrics to a predetermined threshold associated with at least 95% specificity; and applying a machine learning classifier comprising a trained logistic regression model; wherein generating the percentile ranking is performed using both classifiers to generate dual percentile rankings.

[0265] In Example 6, the subject matter of Example 1 includes, detennining a number of the at least one eye-tracking test where the measured eye gaze metrics exceed a predetermined threshold associated with reduced social attention; and determining a positive predictive value for autism spectrum disorder based on the number of tests exceeding the predetermined threshold, wherein exceeding the threshold on three or more tests is associated with a positive predictive value of 100% based on validation in at least 400 children.

[0266] Example 7 is a method for characterizing social attention abilities in a child, comprising: administering to the child a multi-domain eye-tracking test battery comprising at least two eye-tracking tests selected from at least two different domains, wherein the domains comprise: visual social attention, wherein a visual social attention test presents paired social visual stimuli depicting human interaction and non-social visual stimuli; auditory / social attention, wherein an auditory social attention test presents social audio stimuli comprising human speech and non-social audio stimuli; and shifting social attention, wherein a shifting social attention test presents joint attention probes comprising attention-directing gestures; measuring, using an eye-tracking device, eye gaze metrics during each eye-tracking test, the eyeDocket No. 1133.124WO1gaze metrics comprising: fixation data indicating attention allocation between social and non-social stimuli from the visual social attention test; fixation data indicating attention allocation between social audio and non¬ social audio from the auditory social attention test; or attention shifting data from the shifting social attention test; extracting quantitative metrics from the eye gaze metrics for each domain tested; comparing the quantitative metrics to normative data from a plurality of children to determine relative social attention levels; and generating a social attention characterization based on the quantitative metrics from the at least two different domains, wherein the social attention characterization is applicable to children regardless of autism diagnostic status and is used for developmental profiling purposes.|0267| In Example 8, the subject matter of Example 7 includes, wherein the at least two eye-tracking tests comprise at least one visual social attention test, at least one auditory’ social attention test, and at least one shifting social attention test, such that all three domains are assessed.

[0268] In Example 9, the subject matter of Example 7 includes, wherein the auditory’ social attention test comprises gaze-contingent audio stimulus presentation, wherein infant-directed speech characterized by exaggerated inflections and high question rates is activated when the child's gaze is directed to a first area of interest; non-social audio selected from traffic noise or electronic sounds is activated when the child's gaze is directed to a second area of interest; and audio switching occurs with a latency of less than 500 milliseconds based on real-time gaze location monitoring.

[0269] In Example 10, the subject matter of Example 7 includes, wherein the normative data comprises eye gaze metrics from at least 2,000 children aged 12 to 48 months stratified by: age bins of 4-month intervals; and gender; and wherein comparing the quantitative metrics comprises selecting an age-matched and gender-matched comparison group based on the child's age and gender.Docket No. 1133.124WO1|0270] In Example 11, the subject matter of Example 7 includes, wherein the social attention characterization includes: a domain-specific percentile ranking for visual social attention; a domain-specific percentile ranking for auditory social attention; and a domain-specific percentile ranking for shifting social attention; wherein each percentile ranking ranges from 0 to 100 and indicates where the child's performance within each domain falls relative to age-matched and gender-matched children,|02711 In Example 12, the subject matter of Example 7 includes, applying both a simple-threshold classifier and a machine learning classifier to the quantitative metrics; wherein the simple-threshold classifier: compares each quantitative metric to a predetermined threshold empirically derived to achieve at least 95% specificity; and generates a first social attention score; and wherein the machine learning classifier comprises a trained logistic regression model that generates a second social attention score; wherein the social attention characterization comprises percentile rankings for both the first and second social attention scores.

[0272] In Example 13, the subject matter of Example 7 includes, determining a number of eye-tracking tests where the quantitative metrics exceed a predetermined threshold associated with reduced social attention; wherein exceeding the threshold on two tests is associated with a positive predictive value of at least 98% for autism spectrum disorder; and wherein exceeding the threshold on three or more tests is associated with a positive predictive value of 100% for autism spectrum disorder based on validation in at least 400 children who completed at least five eye-tracking tests.

[0273] Example 14 is a method for assessing social attention in a child, the method comprising: administering to the child at least two eye-tracking tests selected from a test battery consisting of: a geometric preference test comprising paired presentation of dynamic geometric images and social images depicting human activity; a complex social test comprising paired presentation of geometric patterns and complex social scenes depicting interpersonal interactions; an outdoor play test comprising pairedDocket No. 1133.124WO1presentation of fractal patterns and social scenes depicting group activities; a first motherese test comprising alternating presentation of infant-directed speech and traffic sounds, wherein presentation is controlled by the child's gaze location; a second motherese test comprising alternating presentation of infant-directed speech and electronic sounds, wherein presentation is controlled by the child's gaze location; and a joint attention test comprising presentation of joint attention probes with attention-directing gestures toward target objects; measuring, using an eye-tracking device, eye gaze metrics during each administered test; extracting quantitative metrics from the measured eye gaze metrics; comparing the extracted quantitative metrics to normative data; and generating a social attention assessment comprising percentile rankings for social attention metrics, wherein the percentile rankings indicate where the child's performance falls relative to the normative data.

[0274] In Example 15, the subject matter of Example 14 includes, wherein the at least two eye-tracking tests comprise at least four tests selected from the test battery, and wherein the social attention assessment is based on performance across the at least four tests.

[0275] In Example 16, the subject matter of Example 14 includes, the geometric preference test comprises a 60 to 65 second movie containing 25 to 30 non-repeating paired dynamic geometric and social image presentations; the complex social test comprises a 90 to 100 second movie containing 8 to 10 paired geometric pattern and complex social scene presentations; the outdoor play test comprises a 70 to 75 second movie containing 9 to 12 paired fractal pattern and social scene presentations; and the joint attention test comprises a 90 to 95 second presentation containing 6 to 10 joint attention probe cycles.

[0276] In Example 17, the subject matter of Example 14 includes, wherein the first motherese test and the second motherese test each utilize gazecontingent technology wherein: real-time gaze coordinate monitoring determines which audio stimulus is presented; infant-directed speech isDocket No. 1133.124WO1characterized by prosodic features comprising elevated pitch, exaggerated intonation, and question intonation patterns; audio switching between infant- directed speech and non-social sounds occurs within 500 milliseconds of gaze location change; and the child's fixation duration on each area of interest determines cumulative exposure time to each audio stimulus type.

[0277] In Example 18, the subject matter of Example 14 includes, extracting quantitative metrics comprises extracting nine distinct metrics comprising: percent fixation on non-social images from visual social attention tests; percent fixation on social images from visual social attention tests; percent fixation on infant-directed speech from the first motherese test and the second motherese test; number of saccades per second within social areas of interest; number of saccades per second within non-social areas of interest; percent fixation on eyes during stimulus presentation; percent fixation on face during stimulus presentation; percent fixation on non-target objects during joint attention probes; and number of complete joint attention cycles comprising target-face -target or face-target-face gaze sequences.

[0278] In Example 19, the subject matter of Example 14 includes, applying a simple-threshold classification approach by: identifying maximum percent fixation on non-social stimuli across all completed tests; determining whether each test exceeds a threshold of 69% fixation on non-social stimuli for visual and auditory tests or 30% fixation on non-target objects for the joint attention test; calculating an autism score equal to the maximum percent fixation value plus 5 points for each additional test exceeding its respective threshold, capped at 100; linearly transforming scores of 69 or higher to a range of 50-100; and applying a machine learning approach by: normalizing features using robust scaling for percentage features and saccade features; inputting normalized features into a trained logistic regression model optimized via 5-fold cross-validation; and generating a machine learning score ranging from 0 to 100; wherein the social attention assessment comprises percentile rankings for both the autism score and the machine learning score.Docket No. 1133.124WO1|0279| In Example 20, the subject matter of Example 19 includes, the normative data comprises eye gaze metrics from at least 2,400 children comprising: children diagnosed with autism spectrum disorder; children with non-autism developmental delays; and typically developing children; the normative data is stratified by 4-month age bins and by gender; the trained logistic regression model achieves: sensitivity of at least 65% on independent test data; specificity of at least 90% on independent test data; and area under the receiver operating characteristic curve of at least 0.85; and model performance has been validated on at least two independent replication cohorts.

[0280] Example 21 is at least one machine -readable medium including instructions that, when executed by processing circuitry, cause the processing circuitry to perform operations to implement of any of Examples 1-20.

[0281] Example 22 is an apparatus comprising means to implement of any of Examples 1-20.

[0282] Example 23 is a system to implement of any of Examples 1-20.

[0283] Example 24 is a method to implement of any of Examples 1-20.

[0284] Examples of the system and method can include every combination and permutation of the various system components and the various method processes, wherein one or more instances of the method and / or processes described herein can be performed asynchronously (e.g., sequentially), concurrently (e.g., in parallel), or in any other suitable order by and / or using one or more instances of the systems, elements, and / or entities described herein.

[0285] As a person skilled in the art will recognize from the previous detailed description and from the figures and claims, modifications andDocket No. 1133.124WO1changes can be made to the examples of the invention disclosed herein without departing from the scope of this invention defined in the following claims.

[0286] The above-detailed description includes references to the accompanying drawings, which form a part of the detailed description. The drawings show, by way of illustration, specific embodiments in which the invention can be practiced. These embodiments are also referred to herein as “examples. ” Such examples can include elements in addition to those shown or described. However, the present inventors also contemplate examples in which only those elements shown or described are provided. Moreover, the present inventors also contemplate examples using any combination or permutation of those elements shown or described (or one or more aspects thereof), either with respect to a particular example (or one or more aspects thereof), or with respect to other examples (or one or more aspects thereof) shown or described herein.

[0287] In the event of inconsistent usages between this document and any documents so incorporated by reference, the usage in this document controls.

[0288] In this document, the terms “a” or “an” are used, as is common in patent documents, to include one or more than one, independent of any other instances or usages of “at least one” or “one or more.” In this document, the term “or” is used to refer to a nonexclusive or, such that “A or B” includes “A but not B,” “B but not A,” and “A and B,” unless otherwise indicated. In this document, the terms “including” and “in which” are used as the plain-English equivalents of the respective terms “comprising” and “wherein.” Also, in the following claims, the terms “including” and “comprising” are open-ended, that is, a system, device, article, composition, formulation, or process that includes elements in addition to those listed after such a term in a claim are still deemed to fall within the scope of that claim. Moreover, in the following claims, the terms “first,” “second,” and “third,” etc. are used merely as labels, and are not intended to impose numerical requirements on their objects.Docket No. 1133.124WO1

[0289] The above description is intended to be illustrative, and not restrictive. For example, the above-described examples (or one or more aspects thereof) may be used in combination with each other. Other embodiments can be used, such as by one of ordinary skill in the art upon reviewing the above description. The Abstract is provided to allow the reader to quickly ascertain the nature of the technical disclosure. It is submitted with the understanding that it will not be used to interpret or limit the scope or meaning of the claims. Also, in the above Detailed Description, various features may be grouped together to streamline the disclosure. This should not be interpreted as intending that an unclaimed disclosed feature is essential to any claim. Rather, inventive subject matter may lie in less than all features of a particular disclosed embodiment. Thus, the following claims are hereby incorporated into the Detailed Description as examples or embodiments, with each claim standing on its own as a separate embodiment, and it is contemplated that such embodiments can be combined with each other in various combinations or permutations. The scope of the invention should be determined with reference to the appended claims, along w ith the full scope of equivalents to which such claims are entitled.

Claims

1. Docket No. 1133.124WO1CLAIMSWhat is Claimed is:

1. A method for assessing social attention in a child, comprising: presenting to a child at least one eye-tracking test comprising visual or auditory stimuli;measuring, using an eye-tracking device, eye gaze metrics during presentation of the visual or auditory stimuli;comparing the measured eye gaze metrics to normative data from a plurality of children; andgenerating a percentile ranking for at least one of the measured eye gaze metrics, wherein the percentile ranking indicates where a social attention level of the child falls relative to the plurality of children.

2. The method of claim 1, wherein the at least one eye-tracking test comprises a multi-domain test battery including at least two tests selected from at least two different domains, the domains comprising:visual social attention tests presenting paired social visual stimuli and non-social visual stimuli;auditory social attention tests presenting social audio stimuli and non-social audio stimuli; andshifting social attention tests presenting joint attention probes.

3. The method of claim 2, wherein at least one auditory social attention test comprises gaze-contingent audio stimulus presentation, wherein infant-directed speech is activated when the child's gaze is directed to a first area of interest and non-social audio is activated when the child's gaze is directed to a second area of interest.

4. The method of claim 1, wherein the normative data comprises eye gaze metrics from at least 2,000 children stratified by age bins of 4-month intervals and by gender, and wherein comparing the measured eye gaze metrics comprises selecting an age-matched and gender-matched comparison group from the normative data.Docket No. 1133.124WO15. The method of claim 1, further comprising:applying a simple-threshold classifier that compares the measured eye gaze metrics to a predetermined threshold associated with at least 95% specificity; andapplying a machine learning classifier comprising a trained logistic regression model;wherein generating the percentile ranking is performed using both classifiers to generate dual percentile rankings.

6. The method of claim 1, further comprising:determining a number of the at least one eye-tracking test where the measured eye gaze metrics exceed a predetermined threshold associated with reduced social attention; anddetermining a positive predictive value for autism spectrum disorder based on the number of tests exceeding the predetermined threshold, wherein exceeding the threshold on three or more tests is associated with a positive predictive value of 100% based on validation in at least 400 children.

7. A method for characterizing social attention abilities in a child, comprising:administering to the child a multi-domain eye-tracking test battery comprising at least two eye-tracking tests selected from at least two different domains, wherein the domains comprise:visual social attention, wherein a visual social attention test presents paired social visual stimuli depicting human interaction and nonsocial visual stimuli;auditory social attention, wherein an auditory social attention test presents social audio stimuli comprising human speech and nonsocial audio stimuli; andshifting social attention, wherein a shifting social attention test presents joint attention probes comprising attention-directing gestures;measuring, using an eye-tracking device, eye gaze metrics during each eye-tracking test, the eye gaze metrics comprising:Docket No. 1133.124WO1fixation data indicating attention allocation between social and non-social stimuli from the visual social attention test;fixation data indicating attention allocation between social audio and non-social audio from the auditory social attention test; or attention shifting data from the shifting social attention test;extracting quantitative metrics from the eye gaze metrics for each domain tested;comparing the quantitative metrics to normative data from a plurality of children to determine relative social attention levels; and generating a social attention characterization based on the quantitative metrics from the at least two different domains, wherein the social attention characterization is applicable to children regardless of autism diagnostic status and is used for developmental profiling purposes.

8. The method of claim 7, wherein the at least two eye -tracking tests comprise at least one visual social attention test, at least one auditory social attention test, and at least one shifting social attention test, such that all three domains are assessed.

9. The method of claim 7, wherein the auditory social attention test comprises gaze-contingent audio stimulus presentation, wherein:infant-directed speech characterized by exaggerated inflections and high question rates is activated when the child's gaze is directed to a first area of interest;non-social audio selected from traffic noise or electronic sounds is activated when the child's gaze is directed to a second area of interest; and audio switching occurs with a latency of less than 500 milliseconds based on real-time gaze location monitoring.

10. The method of claim 7, wherein the normative data comprises eye gaze metrics from at least 2,000 children aged 12 to 48 months stratified by:age bins of 4-month intervals; andgender;Docket No. 1133.124WO1and wherein comparing the quantitative metrics comprises selecting an age-matched and gender-matched comparison group based on the child's age and gender.

11. The method of claim 7, wherein the social attention characterization comprises:a domain-specific percentile ranking for visual social attention; a domain-specific percentile ranking for auditory social attention; anda domain-specific percentile ranking for shifting social attention; wherein each percentile ranking ranges from 0 to 100 and indicates where the child's performance within each domain falls relative to age- matched and gender-matched children.

12. The method of claim 7. further comprising:applying both a simple-threshold classifier and a machine learning classifier to the quantitative metrics;wherein the simple-threshold classifier:compares each quantitative metric to a predetermined threshold empirically derived to achieve at least 95% specificity; and generates a first social attention score; andwherein the machine learning classifier comprises a trained logistic regression model that generates a second social attention score;wherein the social attention characterization comprises percentile rankings for both the first and second social attention scores.

13. The method of claim 7, further comprising:determining a number of eye-tracking tests where the quantitative metrics exceed a predetermined threshold associated with reduced social attention;wherein exceeding the threshold on two tests is associated with a positive predictive value of at least 98% for autism spectrum disorder; and wherein exceeding the threshold on three or more tests is associated with a positive predictive value of 100% for autism spectrum disorderDocket No. 1133.124WO1based on validation in at least 400 children who completed at least five eyetracking tests.

14. A method for assessing social attention in a child, the method comprising:administering to the child at least two eye-tracking tests selected from a test battery consisting of:a geometric preference test comprising paired presentation of dynamic geometric images and social images depicting human activity;a complex social test comprising paired presentation of geometric patterns and complex social scenes depicting interpersonal interactions;an outdoor play test comprising paired presentation of fractal patterns and social scenes depicting group activities;a first motherese test comprising alternating presentation of infant-directed speech and traffic sounds, wherein presentation is controlled by the child's gaze location;a second motherese test comprising alternating presentation of infant-directed speech and electronic sounds, wherein presentation is controlled by the child's gaze location; anda joint attention test comprising presentation of joint attention probes with attention-directing gestures toward target objects;measuring, using an eye-tracking device, eye gaze metrics during each administered test;extracting quantitative metrics from the measured eye gaze metrics;comparing the extracted quantitative metrics to normative data; andgenerating a social attention assessment comprising percentile rankings for social attention metrics, wherein the percentile rankings indicate where the child's performance falls relative to the normative data.Docket No. 1133.124WO115. The method of claim 14, wherein the at least two eye-tracking tests comprise at least four tests selected from the test battery, and wherein the social attention assessment is based on performance across the at least four tests.

16. The method of claim 14, wherein:the geometric preference test comprises a 60 to 65 second movie containing 25 to 30 non-repeating paired dynamic geometric and social image presentations;the complex social test comprises a 90 to 100 second movie containing 8 to 10 paired geometric pattern and complex social scene presentations;the outdoor play test comprises a 70 to 75 second movie containing 9 to 12 paired fractal pattern and social scene presentations; and the joint attention test comprises a 90 to 95 second presentation containing 6 to 10 joint attention probe cycles.

17. The method of claim 14, wherein the first motherese test and the second motherese test each utilize gaze-contingent technology wherein:real-time gaze coordinate monitoring determines which audio stimulus is presented;infant-directed speech is characterized by prosodic features comprising elevated pitch, exaggerated intonation, and question intonation patterns;audio switching between infant-directed speech and non-social sounds occurs within 500 milliseconds of gaze location change; andthe child's fixation duration on each area of interest determines cumulative exposure time to each audio stimulus type.

18. The method of claim 14, wherein extracting quantitative metrics comprises extracting nine distinct metrics comprising:percent fixation on non-social images from visual social attention tests;percent fixation on social images from visual social attention tests;Docket No. 1133.124WO1percent fixation on infant-directed speech from the first motherese test and the second motherese test;number of saccades per second within social areas of interest; number of saccades per second within non-social areas of interest;percent fixation on eyes during stimulus presentation; percent fixation on face during stimulus presentation; percent fixation on non-target objects during joint attention probes; andnumber of complete joint attention cycles comprising target-face-target or face -target-face gaze sequences.

19. The method of claim 14, further comprising:applying a simple-threshold classification approach by:identifying maximum percent fixation on non-social stimuli across all completed tests;determining whether each test exceeds a threshold of 69% fixation on non-social stimuli for visual and auditory tests or 30% fixation on non-target objects for the joint attention test;calculating an autism score equal to the maximum percent fixation value plus 5 points for each additional test exceeding its respective threshold, capped at 100;linearly transforming scores of 69 or higher to a range of 50-100; andapplying a machine learning approach by:normalizing features using robust scaling for percentage features and saccade features;inputting normalized features into a trained logistic regression model optimized via 5-fold cross-validation; andgenerating a machine learning score ranging from 0 to 100;wherein the social attention assessment comprises percentile rankings for both the autism score and the machine learning score.Docket No. 1133.124WO120. The method of claim 19, wherein:the normative data comprises eye gaze metrics from at least 2,400 children comprising:children diagnosed with autism spectrum disorder; children with non-autism developmental delays; and typically developing children;the normative data is stratified by 4-month age bins and by gender;the trained logistic regression model achieves:sensitivity of at least 65% on independent test data; specificity of at least 90% on independent test data; and area under the receiver operating characteristic curve of at least 0.85; andmodel performance has been validated on at least two independent replication cohorts.

Images