System and method for detecting neurodevelopmental conditions

Abstract

Disclosed are techniques for detecting a neurodevelopmental condition. A system may be provided that includes processing unit(s) configured to perform various tasks. Such tasks may include causing an auditory stimulus to be experienced by a user, receiving audio and video of the user that includes experiencing the auditory stimuli and for a period of time afterwards, making auditory and visual assessments of the captured responses to the auditory stimuli, and determining a neurodevelopmental condition is present based on the auditory assessment and / or the visual assessment.

Description

[0001] SYSTEM AND METHOD FOR DETECTING NEURODEVELOPMENTAL CONDITIONS

[0002] CROSS-REFERENCE TO RELATED APPLICATIONS

[0003] The present application claims priority to U.S. Provisional Patent Application No. 63 / 548,582, filed February 1, 2024, the contents of which are incorporated by reference herein in its entirety.

[0004] TECHNICAL FIELD

[0005] The present disclosure is drawn towards techniques for detecting neurodevelopmental conditions, such as autism spectrum disorder (ASD).

[0006] BACKGROUND

[0007] This section is intended to introduce the reader to various aspects of art, which may be related to various aspects of the present invention that are described and / or claimed below. This discussion is believed to be helpful in providing the reader with background information to facilitate a better understanding of the various aspects of the present invention. Accordingly, it should be understood that these statements are to be read in this light, and not as admissions of prior art.

[0008] Assessing behavioral responses to stimuli, particularly unstructured behavioral responses such as vocalizations, head movements, and oral-facial gestures, is critical in understanding a wide spectrum of neurodevelopmental conditions. The variability in vocalization patterns, both in terms of acoustic and phonological features, has been associated with, e.g., autism symptom seventy. Similarly, atypical head movements and the dynamics of these movements have been observed as distinguishing characteristics in individuals with autism. Recent studies have identified that oral-facial gestures, such as patterns of mouth opening and closing, are also associated with other neurodevelopmental conditions, including Attention-Deficit Hyperactivity Disorder (ADHD).

[0009] However, existing systems and methods for capturing and analyzing and quantifying these unstructured behavioral responses are often limited in scope and application.

[0010] BRIEF SUMMARY

[0011] Various deficiencies in the prior art are addressed below by the disclosed compositions of matter and techniques. In various aspects, a system for detecting a neurodevelopmental condition may be provided. The system may include one or more processing units configured to, collectively, perform several tasks. The tasks may include causing an auditory stimulus pattern to be experienced by a user (e.g, causing an auditory stimulus pattern to be generated in a way that a user could experience it). The tasks may include receiving audio and video of the user for a first period of time. The first period of time may include time prior to the user experiencing the auditory stimuli. The first period of time may include a time during which the user experienced the auditory stimuli. The first period of time may include a second period of time after the user experienced the audit ory stimuli. The tasks may include making an auditory assessment by determining if the audio includes syllabic vocalizations, non-syllabic vocalizations, or both, in response to an auditory stimulus pattern. The tasks may include making a visual assessment by determining at least two spontaneous unstructured movements in response to the auditory stimuli, the at least two spontaneous unstructured movements including a degree of openness of at least one eye and at least one additional spontaneous unstructured movement. The tasks may include determining a neurodevelopmental condition (such as ASD and / or ADHD) is present based on the auditory assessment and / or the visual assessment.

[0012] The at least one additional spontaneous unstructured movement may include a degree of openness of a mouth, a degree of head movement, an arm or hand movement, or a combination thereof. The arm or hand movement may be determined to be the user reaching for an ear or the user covering an ear.

[0013] The auditory stimuli may be white noise. The auditory stimuli may include one or more discrete frequencies of sound. The discrete frequencies of sound may be played simultaneously (e.g., 200 Hz, 8000 Hz, and 16000 Hz played simultaneously), or sequentially.

[0014] The second period of time may be no more than 5 seconds. The second period of time may be no more than 2 seconds.

[0015] The neurodevelopmental condition may be determined based solely on the auditory assessment. The neurodevelopmental condition may be determined based solely on the visual assessment. The neurodevelopmental condition may be determined based solely on the combination of the audio assessment and the visual assessment.

[0016] The method may include using mouth state data (e.g., open vs closed, or degree of openness from 0-100%, etc.) in conjunction with at least one other behavioral indicator to assess the neurodevelopmental condition. The auditory assessment may be informed by research findings on vocalization patterns in individuals with the neurodevelopmental condition. The visual assessment may be informed by research findings on head movements in individuals with the neurodevelopmental condition.

[0017] The system may include a first display. The system may include a camera. The system may include a speaker. The system may include a memory . The processing unit(s) may be operably coupled to a display, camera, speaker, and memory. The system may be configured as a desktop computer, a laptop computer, a mobile phone, or a tablet.

[0018] The one or more processing units may include a local processor (e.g., a processor on a mobile phone) and a remote processor (e.g., a processor of a cloud-based server, or a computer at a healthcare facility). In some embodiments, the remote processor may be configured to perform the auditory and visual assessment steps.

[0019] The display and camera may be operably coupled to a headset, the headset being operably coupled to the one or more processing units. The speakers may be coupled to the headset. The speakers may be configured as headphones.

[0020] In various aspects, anon-transitory computer-readable storage device may be provided. The storage device may contain instructions that, when executed by one or more processing units, causes the one or more processing units to, collectively, perform a method, such as a method for measuring unstructured behavioral responses may be provided. The method may utilize an embodiment of a sy stem as disclosed herein. The method may include exposing an individual to auditory' stimuli. The method may include recording any vocalizations, head movements, and / or mouth movements of the individual, and particularly those in response to the auditory stimuli. The method may include identifying one or more patterns by analyzing the vocalizations, head movements, and / or mouth movements. The method may include determining if the one or more patterns are associated with a neurodevelopmental condition. The method may include providing an assessment. The assessment may be provided to a clinician or therapist.

[0021] In various aspects, a method for measuring unstructured behavioral responses may be provided. The method may utilize a system as disclosed herein. The method may include receiving an assessment. The assessment may have been generated for an individual utilizing a computer-based assessment system as disclosed herein. The computer-based assessment system may be configured to expose an individual to auditory stimulus, record any vocalizations, head movements, and / or mouth movements of the individual, identify one or more patterns by analyzing the vocalizations, head movements, and / or mouth movements, determine if the one or more patterns are associated with a neurodevelopmental condition, and provide an assessment as disclosed herein. The method may include, based on the assessment, instituting or adjusting therapy for the individual responsive to the neurodev elopmental condition. The method may include instructing the individual to utilize the computer-based assessment system.

[0022] BRIEF DESCRIPTION OF DRAWINGS

[0023] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments of the present invention and, together with a general description of the invention given above, and the detailed description of the embodiments given below, serve to explain the principles of the present invention.

[0024] Figure 1A is a diagram of an embodiment of a system.

[0025] Figure IB is an illustration of an embodiment of a headset.

[0026] Figure 2A is a schematic of an embodiment of a technique for detecting neurodevelopmental conditions.

[0027] Figure 2B is a schematic showing data collection using a disclosed smartphone-based platform for neurobehavioral evaluations and analysis using machine learning algorithms to assess the diagnostic accuracy for ASD.

[0028] Figures 3A-3F show box plots and density plots illustrating increased response levels in children with ASD (solid line) compared to neurotypical (NT) children (dotted line) for screen avoidance (3A), Anteroposterior (AP) postural stability (3B), head rotations (3C), headphone touches (3D), mouth openings (3E), and non-syllabic vocalizations (3F).

[0029] Figure 3G shows graphs illustrating group averaged profiles of startle eye blinks (normalized eyelid closure, NEC), mouth openings, non-syllabic vocalizations (NSV), and laughter around the onset of the auditory stimulus in children with ASD (solid line) compared to NT children (dotted line).

[0030] Figure 4 is a table showing demographic and clinical characteristics of participants in a study, where ASD = Autism Spectrum Disorder and SD = Standard Deviation,

[0031] Figure 5 is an example flowchart of sequential data processing and machine learning analysis steps, beginning with the import of raw data and continuing through multiple stages of a PyCaret machine learning pipeline.

[0032] Figure 6 is a table containing smartphone-based assessments of behavioral responses in children with ASD compared to neurotypical children.

[0033] Figures 7A-7I are plots showing characteristics of the auditory startle response, longterm startle habituation, and prepulse inhibition in children with autism (ASD) and neurotypical (NT) children. In these figures, significance levels of *=p < 0.05, **=p < 0.01, and ***=p < 0.001 may be shown. For Figures 7A-7C and 7E, the vertical dashed lines indicates the onset of either an auditory prepulse or onset of the pulse as appropriate. Gray vertical columns indicate the stimulus durations of 50 ms. In particular:

[0034] Figure 7A is a plot showing group averaged eyelid movement plots showing the acoustically-evoked startle responses for girls (left panel) and boys (right panel) expressed as the amplitude of normalized eyelid closure (NEC); note that in ASD girls the magnitude of the startle response is marginally larger than in ASD boys, however no significant sex * diagnosis interaction effects were found.

[0035] Figure 7B is a plot showing, at group level, ASD (blue) and NT (orange) children had comparable eyelid startle responses.

[0036] Figure 7C is a plot showing group averaged eyelid movement plots of the pulse-only trials ordered by trial number. Trials 1-2 were delivered at the start and trials 9-10 at the end of the 15-minute smartphone test.

[0037] Figure 7D is a plot showing mean eyelid startle amplitude as a function of trial number (#) for ASD and NT children; Note that ASD and NT groups have comparable levels of longterm startle habituation.

[0038] Figure 7E is a plot showing group averaged eyelid movement plots ordered by stimulus type, i.e. by the intensity of the prepulse that preceded the pulse.

[0039] Figure 7F is a plot showing mean amplitude of the eyelid startle response as a function of prepulse intensity; note that the group with NT children exhibited a typical prepulse inhibition (PPI) response, characterized by a decrease in mean eyelid startle amplitude as the intensity of the prepulse increased. The group of children diagnosed with ASD did not display evidence of PPI.

[0040] Figure 7G is a plot showing variability of the amplitude of the eyelid startle responses as a function of prepulse intensity for ASD and NT children; For all prepulse intensities the group of children with ASD had a higher variability than NT children.

[0041] Figure 7H are box and density plots showing the amplitude of the eyelid startle response during trials with a high (25%) prepulse intensity relative to the amplitude in pulse only trials (indicated with the horizontal dashed line at y-value of 100); note that in more than 50 percent of the trials there was no decrease, but even an increase, in the startle response amplitude in children with ASD. Figure 71 are box and density plots showing Increased variability in the amplitude of the eyelid startle response during trials with a high prepulse intensity in children with ASD compared to NT children.

[0042] Figures 8A-8F are plots relating to short-term startle habituation and anticipatory eyeblinks in a rhythmic and random stimulus pattern. In these figures, significance levels of *=p < 0.05, **=p < 0.01, and ***=p < 0.001 may be shown. In particular:

[0043] Figures 8A and 8B are group averaged plots of eyelid movement over a duration of five seconds during which eyelid startle responses were evoked by the presentation of six white noise pulses in either a rhythmic (8 A) or a more random (8B) stimulus pattern. Each 15-minute smartphone experiment consisted of ten rhythmic and ten random trials. Onset of the white noise pulses is indicated with the vertical dashed lines, duration of pulses (50 ms each) by the gray vertical columns. The magnitude of the startle pulse-evoked eyelid movements, expressed as normalized eyelid closure (NEC), was quantified in the ‘startle windows’ indicated by the yellow shaded areas SI to S6. The magnitude of anticipatory eyeblink (AEB) responses around the moment of omitted stimuli was quantified in the gray-shaded observation windows (OW) 1 and 2. Note that NT children showed no short-term habituation and that children with ASD even showed a sensitization (i.e. startle responses becoming larger after the first pulse). In addition, children with ASD showed stronger AEB responses than NT children.

[0044] Figures 8C and 8D are group average plots showing the cumulative sum of eyelid movements shown in Figures 8A and 8B, respectively. Note the increased cumulative sum in children with ASD compared to NT children during the random trials.

[0045] Figures 8E and 8F are box and density plots representing the magnitude of the eyelid startle responses in the marked observation windows (“[E]” and “[F]”) in Figures 8A and 8B.

[0046] Figure 9A is a matrix visualizing the correlation coefficients between various neurobehavioral markers collected during prepulse inhibition (PPI) and habituation (HAB) protocols from children with Autism Spectrum Disorder (ASD) compared to neurotypical (NT) controls. Heat represents the strength of association, with 1 indicating perfect correlation and 0 no correlation.

[0047] Figure 9B is a plot showing performance of the predictive model on the holdout set (30% of dataset) evaluated using Receiver Operating Characteristic (ROC) curves for ASD (blue line) and NT (orange line) classifications. The graph plots true positive rates against false positive rates at various decision threshold levels. The table presents the confusion matrix, with values representing the count of correct and incorrect classifications. Figure 9C is a plot showing feature importance scatter plot generated by Shapley additive explanation (SHAP) values, ranking the influence of specific neurobehavioral features on the predictive models (AP = Anteroposterior).

[0048] Figure 10 is a table showing stratified 10 fold cross-validation scores per feature set in random forest classifier plus the holdout scores for the final model. Feature set performances in a random forest classifier for ASD prediction ranked by accuracy. Metrics are obtained by training a model on only the selected set of features. For the relative importance of features in our final model we refer to SHAP values reported in figure 5. AEB = Anticipatory eye blink AUG = Area Under the Curve, HO = Holdout set, NPV = Negative Predictive Value, PPV = Positive Predictive Value, AP = Anterior Posterior, ANT = Anticipation, PPI = Prepulse Inhibition, CV = Cross Validation.

[0049] It should be understood that the appended drawings are not necessarily to scale, presenting a somewhat simplified representation of various features illustrative of the basic principles of the invention. The specific design features of the sequence of operations as disclosed herein, including, for example, specific dimensions, orientations, locations, and shapes of various illustrated components, will be determined in part by the particular intended application and use environment. Certain features of the illustrated embodiments have been enlarged or distorted relative to others to facilitate visualization and clear understanding. In particular, thin features may be thickened, for example, for clarity or illustration.

[0050] DETAILED DESCRIPTION

[0051] Between 69% and 95% of children with autism spectrum disorder (ASD) experience sensory difficulties, frequently characterized by hypo- or hypersensitivity in multiple sensory modalities, including auditory and encompassing more variable response patterns indicative of noisy processing. Recently, the largest population-based study of over 25,000 children with ASD found that this altered sensitivity is strongly associated with other ASD symptoms, such as difficulties in adaptive behavior, emotional states, aggression, attention, fear, and motor development, but not with global intellectual ability.

[0052] In clinical practice, sensory sensitivity is commonly evaluated through questionnaires such as the Sensory Profile (Dunn, 1999) and the Sensory Processing Measure (Parham et al., 2007). These questionnaires often rely on subjective observations and interpretations, which can be influenced by observer bias. In addition, the diagnostic evaluations process is timeconsuming thus contributing to the autism waitlist crisis. As autism research has primarily focused on high-income Western populations, these questionnaire-based assessments might have geographic and socio-economic biases. Sensory difficulties are less likely to be detected in children belonging to ethnic minority groups in the United States compared to White, nonHispanic children (odds ratios: 0.71-0.78), which may stem from disparities in identification and access to specialized services. Additionally, the odds of reporting sensory features was 0.83 times lower in girls than boys.

[0053] A more objective methodology to quantify sensory sensitivity involves the use of neurobehavioral evaluations, which have proven valuable in pre-clinical autism research and often can differentiate children with ASD from neurotypical children. These assessments include spontaneous and stimulus-evoked general behaviors, such as head-and-body movements and vocalizations, and specific neurometric examinations, such as the acoustically evoked eyelid startle reflex (ASR) and tests that involve the modulation of the ASR, including prepulse inhibition (PPI) and habituation (HAB).

[0054] However, detailed neurobehavioral assessments are not part of routine clinical practice due to technical complexities associated with their administration.

[0055] A user-friendly smartphone-based platform has been developed specifically designed for conducting neurobehavioral evaluations. This platform, called BlinkLab, is optimized for at-home use and yields robust and reproducible results across various neurometric tests, providing the opportunity to quantify spontaneous and stimulus-evoked behaviors.

[0056] Disclosed herein are techniques where smartphone-based neurobehavioral assessments focusing on sensory sensitivity alone can serve as an accurate diagnostic tool for ASD.

[0057] The following description and drawings merely illustrate the principles of the invention. It will thus be appreciated that those skilled in the art will be able to devise various arrangements that, although not explicitly described or shown herein, embody the principles of the invention and are included within its scope. Furthermore, all examples recited herein are principally intended expressly to be only for illustrative purposes to aid the reader in understanding the principles of the invention and the concepts contributed by the inventor(s) to furthering the art and are to be construed as being without limitation to such specifically recited examples and conditions. Additionally, the term, "or," as used herein, refers to a nonexclusive or, unless otherwise indicated (e.g., “or else” or “or in the alternative”). Also, the various embodiments described herein are not necessarily mutually exclusive, as some embodiments can be combined with one or more other embodiments to form new embodiments.

[0058] The numerous innovative teachings of the present application will be described with particular reference to the presently preferred exemplary embodiments. However, it should be understood that this class of embodiments provides only a few examples of the many advantageous uses of the innovative teachings herein. In general, statements made in the specification of the present application do not necessarily limit any of the various claimed inventions. Moreover, some statements may apply to some inventive features but not to others. Those skilled in the art and informed by the teachings herein will realize that the invention is also applicable to various other technical areas or embodiments.

[0059] Disclosed is a comprehensive and versatile system that can objectively measure these responses across various neurodevelopmental conditions, providing valuable insights for both clinical assessments and therapeutic interventions.

[0060] The present disclosed overcomes various deficiencies in conventional approaches by, inter alia, providing a method and system for objectively measuring unstructured behavioral responses (e.g., spontaneous and stimulus-evoked behavioral responses), including vocalizations, head movements and oral-facial gestures. For measuring the vocalizations, the system may use a smartphone or tablet’s microphone. For capturing the oral-facial gestures, the system may be computer vision algorithms to detect whether the mouth is open or closed and measure the area and degree of openness with precision during the neurobehavioral test. The system may employ advanced sensors and analytical tools to capture and interpret these responses, revealing patterns and associations with neurodevelopmental conditions.

[0061] The system may be equipped with audio capturing devices to record vocalizations, distinguishing between syllabic and nonsyllabic vocalizations. The system's analytical capabilities may be informed by research indicating the significance of nonsyllabic vocalizations and the ratio of speech to non-speech vocalizations in identifying and assessing ASD.

[0062] In various aspects, a system for detecting a neurodevelopmental condition may be provided. Referring to FIG. 1A, a system (100) may include one or more processing units (110) configured to, collectively, perform several tasks in order to detect a neurodevelopmental condition.

[0063] As used herein, The term “processing unit” (or “processor”) includes any combination of hardware, firmware, and software, employed to process data or digital signals. Processing unit hardware may include, for example, application specific integrated circuits (ASICs), general purpose or special purpose central processing units (CPUs), digital signal processors (DSPs), graphics processing units (GPUs), and programmable logic devices such as field programmable gate arrays (FPGAs). The system may include a memory (112) operably coupled to one or more processing unit(s). The system may include a non-transitory computer-readable storage device (114) operably coupled to one or more processing unit(s). The system may include an I / O interface (such as a wireless communications interface) (116) operably coupled to one or more processing unit(s).

[0064] The system may include a speaker (120). In some embodiments, the speaker may be a set of headphones (162) or earphones. Such headphones or earphones may be placed in or on the head of a user (160). Such headphones or earphones may be operably coupled, either wired or wirelessly, to processing unit(s). For example, FIG. 1A shows a mobile phone (102) with a headphone jack (132), where the headphones (162) are connected to the mobile phone through the headphone jack. The processing unit(s) may be operably coupled to the speaker. The system may include a microphone (130). The processing unit(s) may be operably coupled to the microphone. The system may include a camera (140). The processing unit(s) may be operably coupled to the camera. The sy stem may include a first display (150). The processing unit(s) may be operably coupled to the first display.

[0065] The system may be configured in any appropriate form. For example, the system may be configured as a desktop computer, a laptop computer, a mobile phone, or a tablet.

[0066] The one or more processing units may include a local processor (e.g., processor (111) on a mobile phone (102)), and a remote processor (e.g., processor (171) of a cloud-based server (170)), or a computer at a healthcare facility). Which processor or processing unit performs which tasks may be readily adjusted, as understood by those of skill in the art. In some embodiments, the remote processor may be configured to perform auditory and visual assessment steps, while the local processor performs steps needed for eliciting and capturing reactions to various stimuli.

[0067] The processing umt(s) may include a remote processor (173) associated with a researcher or clinician (for example, a processor found at ahealthcare facility (172) or research facility). That processor may be configured to receive input from, or provide output to, a clinician or researcher. In some embodiments, any remote processor may be configured to interact with a local processor (111). In some embodiments, only a cloud-based remote processor (171) may be configured to communicate with the local processor (111), and any outside user (such as a clinician or researcher) would be configured to communicate with that cloud-based remote processor, rather than directly with the local processor.

[0068] Referring to FIG. IB, a display (150) may be operably coupled to a headset (180). A camera (140) may be operably coupled to a headset. A microphone (130) may be operably coupled to a headset. Speakers (120) may be operably coupled to a headset. The speakers may be configured as headphones. It is understood that while FIG. IB shows the speakers as built- in headphones, the speakers could also be separate headphones placed over the user’s ears.

[0069] The headset may be operably coupled to the one or more processing units. In some embodiments, the headset may include a processing unit (181). In some embodiments, the headset is coupled (e.g., wired or wirelessly) to a local device that includes one or more processing unit(s) (see, e.g., the device of FIG. 1A).

[0070] As noted previously, the one or more processing unit(s) (110) may be configured to perform specific tasks. In some embodiments, a non-transitory computer readable storage device (114) may include instructions that, when executed by the processing unit(s), configures the processing unit(s) to perform the various tasks.

[0071] The tasks may include causing an auditory stimulus pattern to be experienced by a user (e.g. , causing one or more auditory stimuli to be generated in a way that a user could experience it). This may include showing video to a user, and then at one or more times during the video, causing an auditory stimulus pattern to be generated. The times the auditory stimuli are generated may be predetermined, or may be randomly occurring.

[0072] The auditory stimuli may include one or more pulses of sound. The length of time in which each pulse is generated is generally quite short, and may be no more than 250 ms, no more than 200 ms, no more than 150 ms, no more than 100 ms, or no more than 50 ms. Each pulse may be at least 1 ms, at least 5 ms, at least 10 ms, or at least 25 ms. The auditory stimuli may include a “prepulse” with a first intensity followed by a pulse with a second intensity greater than the first intensity.

[0073] The auditory stimuli may be white noise. The auditory stimuli may include one or more discrete frequencies of sound. The discrete frequencies of sound may be played simultaneously (e.g., 200 Hz, 8000 Hz, and 16000 Hz played simultaneously), or sequentially.

[0074] The tasks may include receiving audio and video of the user for a first period of time. That first period of time may preferably include a time prior to the user’s exposure to one or more auditory stimuli. The first period of time may also preferably include a time during which the user experiences the auditory stimuli. The first period must include a second period of time after the user experiences the auditory stimuli.

[0075] The second period of time may be any appropriate amount of time. For example, the second period of time may be no more than 5 minutes. The second period of time may be no more than 2 minutes. The second period of time may be no more than 1 minute. The second period of time may be no more than 30 seconds. The second period of time may be no more than 20 seconds. The second period of time may be no more than 15 seconds. The second period of time may be no more than 10 seconds. The second period of time may preferably be no more than 5 seconds. The second period of time may more preferably be no more than 2 seconds.

[0076] The first period of time will necessarily at least as long as the second period of time, and preferably will be longer. The first period of time may be any appropriate amount of time. For example, the first period of time may be no more 10 minutes longer, 5 minutes longer, 2 minutes longer, 1 minute longer, 30 seconds longer, or 15 seconds longer than the second period of time. The first period of time may preferably be no more than 10 seconds longer than the second period of time. The first period of time may be less than 5 seconds longer than the second period of time. The first period of time may be less than 2 seconds longer than the second period of time. The first period of time may be exactly the same as the second period of time.

[0077] The tasks may include making an auditory assessment by determining if the audio includes syllabic vocalizations, non-syllabic vocalizations, or both, in response to an auditory stimuli.

[0078] The tasks may include making a visual assessment by determining at least two spontaneous unstructured movements in response to the auditory stimuli, the at least two spontaneous unstructured movements including a degree of openness of at least one eye and at least one additional spontaneous unstructured movement. The at least one additional spontaneous unstructured movement may include a degree of openness of a mouth, a degree of head movement, an arm or hand movement, or a combination thereof. The arm or hand movement may be determined to be the user reaching for an ear or the user covering an ear.

[0079] As will be understood, these systems may be implemented in various manners. For example, in some embodiments, a system as disclosed herein may be used as a tool or a clinician or therapist. For example, a method may include instructing an individual to utilize the computer-based assessment system to assess a neurodevelopmental condition. As one example, a therapist may request parents have a child take a test at home using an application on a smartphone and send the results to the therapist. Alternatively, the clinician or therapist may have the test performed in a more controlled environment. The clinician or therapist may receive an assessment from the system, and may, based on the assessment, institute or adjust therapy for the individual responsive to the neurodevelopmental condition.

[0080] Example 1 Referring to FIG. 2A, data were collected using a smartphone-based platform for neurobehavioral evaluations and analyzed using machine learning algorithms to assess the diagnostic accuracy for ASD. Children watched a 15-minute video on a smartphone during which brief auditory stimulus patterns were delivered. The smartphone’s camera captured the child’s postural, head, facial, and vocal responses before and after stimulus onset. Conventional techniques for tracking body part locations and actions were utilized. The illustrations shown represent a subset of tracked responses.

[0081] Referring to FIG. 2B, data were preprocessed to create subject-level behavioral sensory profiles. Group-level statistical analyses revealed significant disparities in sensory responses between children diagnosed with ASD and neurotypical children. Binary classification was done with a random forest algorithm. Note that comparable results were achieved in the cross- validation (CV) set and the holdout set in terms of sensitivity and specificity, indicating that the model generalizes well to unseen data. Values shown in the confusion matrices are the row normalized percentages indicating sensitivity (true positive box) and specificity (true negative box).

[0082] Referring to FIG. 3 A-3F, results of the process can be seen. Here, postural, head, facial, and vocal responses during the smartphone neurobehavioral evaluations can distinguish children with autism (ASD) from neurotypical (NT) children. In FIGS. 3A-3F, box plots and density plots illustrating increased response levels in children with ASD (blue) compared to NT children (orange) are shown. Anteroposterior (AP) postural stability and mouth openings are presented in arbitrary units (a.u.). For the boxplots, visibility within the main body of the data was optimized by setting the y-axis limits to the 10th and 90th percentiles. A missing cap at the whiskers of the box plot means that the whiskers are not capturing the full data range. Note that range in density plots can therefore extend the range in the box plot.

[0083] In FIG. 3G, group averaged profiles of startle eye blinks (normalized eyelid closure, NEC), mouth openings, non-syllabic vocalizations (NSV), and laughter are shown around the onset of the auditory stimulus. Left vertical dash line indicates the onset of the auditory prepulse, the right one indicates the onset of the pulse. Gray vertical columns indicate the stimulus durations of 50 ms. Note that the auditory stimulus evokes a startle eyeblink with a short latency to onset, followed by a mouth opening and vocalization with a longer latency to onset. Eye blink traces represent the trials with the high prepulse intensity at 25% of the pulse intensity. All traces show the difference compared to the baseline prior to stimulus onset (i.e. baselines were aligned at zero). Blue and orange shadings indicate the 95% confidence intervals. Significance levels: * p < 0.05, ** p < 0.01, *** p < 0.001. Thus, it will be understood that models can be trained utilizing postural, head, facial, and vocal responses captured by a camera before and after stimulus onset, with data from subjects that are neurotypical and subjects that are diagnosed with a target neurodev el opmental condition (such as ASD, ADHD, etc.). These trained models can distinguish between individuals considered neurotypical, and those individuals with a neurodevelopmental condition.

[0084] The tasks of the processing unit(s) may include determining a neurodevelopmental condition (such as ASD and / or ADHD) is present based on the auditory assessment and / or the visual assessment. This may include considering if or when a particular type of response occurs after an auditor}' stimuli (e.g., laughter, nonsyllabic vocalizations, etc.), or the degree to which a particular response is detected (e.g., how much eyelid closure is detected).

[0085] The neurodevelopmental condition may be determined based solely on the auditory assessment. The neurodevelopmental condition may be determined based solely on the visual assessment. The neurodevelopmental condition may be determined based solely on the combination of the audio assessment and the visual assessment.

[0086] The method may include using mouth state data e.g., open vs closed, or degree of openness from 0-100%, etc.) in conjunction with at least one other behavioral indicator to assess the neurodevelopmental condition.

[0087] The auditory assessment may be informed by research findings on vocalization patterns in individuals with the neurodevelopmental condition. The visual assessment may be informed by research findings on head movements in individuals with the neurodevelopmental condition.

[0088] In various aspects, a method for measuring unstructured behavioral responses may be provided. The method may utilize an embodiment of a system as disclosed herein. The method may include exposing an individual to auditory stimuli. The method may include recording any vocalizations, head movements, and / or mouth movements of the individual, and particularly those in response to the auditory stimuli. The method may include identifying one or more patterns by analyzing the vocalizations, head movements, and / or mouth movements. The method may include determining if the one or more patterns are associated with a neurodevelopmental condition. The method may include providing an assessment. The assessment may be provided to a clinician or therapist.

[0089] Example The BlinkLab tests were administered to children, both with and without ASD diagnoses. Utilizing machine learning algorithms on neurobehavioral assessments conducted via BlinkLab, clinical assessments for individuals with ASD were confirmed.

[0090] Participants

[0091] A cohort of 280 participants was recruited. Children diagnosed with ASD, comprising 57 girls and 126 boys, were recruited at the Mohammed VI National Center for the Disabled (CNMH) at eight locations in Morocco, including Fes, Sale, Safi, Marrakesh, Casablanca, Oujda, Tangier, and Agadir. The autism diagnosis was established by a multidisciplinary team of specialists using the Diagnostic and Statistical Manual of Mental Disorders (DSM-5; American Psychiatric Association, 2013) criteria, prior to recruitment. Additionally, neurotypical controls, consisting of 43 girls and 54 boys without a neuropsychiatric diagnosis, were recruited from two schools, one in Taounate (90 km from Fes) and the other in Sale. Participants were selected regardless of sex, gender identity or race. Excluded were participants under 3 or over 12 years old, as well as those using medication that affects the nervous system (classified as ATC NO medication, www.whocc.no). See FIG. 4.

[0092] Experimental setup

[0093] Neurobehavioral testing was performed using BlinkLab, a smartphone-based platform. The tests included general measurement of spontaneous and stimulus-evoked postural, head, facial, and vocal responses along with specific neurometric tests, including the ASR, PPI, longterm HAB, and short-term HAB. Children participated in two consecutive 15-minute tests. See, e.g., FIGS. 2A-2B.

[0094] During the experiment, the children watched an audio-normalized movie while the trials containing the auditory stimuli were delivered via headphones (e.g., wired headphones). For each trial, computer vision algorithms were used to track and record the position of the participant’s facial landmarks over time, and to determine amplitude and timing of mouth opening, head and postural movements, as well as the eyelid closure (see FIG. 2A). Eyelid position signals were captured and calculated in real time as previously described (Boele et al., 2023).

[0095] Outcome measures

[0096] For behavioral analysis, various items were quantified: 1. The percentage of trials wherein the child was avoiding the smartphone’s screen and camera. 2. The child’s anteroposterior postural stability. 3. The percentage of trials wherein the child was touching the headphone. 4. The percentage of trials wherein the child was rotating their head. 5. The size and timing of the child’s mouth openings and closings. 6. The percentage of trials wherein the child was vocalizing and the corresponding time of occurrence.

[0097] For the analysis of the neurometric data, the same data analysis pipeline was used as described previously by Boele and colleagues (2023). In addition, for the random and rhythmic stimulus patterns in the short-term HAB protocol, the anticipatory eye blinks (AEB) in predefined observation windows (OW) containing an omitted pulse were analyzed. AEB were responses that were not triggered by the startle stimulus itself but rather by the expectation of an upcoming startle stimulus. For all reflexive and anticipatory blinks, the subject average and variability (standard deviation) were calculated.

[0098] Prepulse Inhibition (PPI) paradigm'. We studied PPI of the acoustically evoked eyelid startle response, using a 50 ms white noise audio burst at 105 dB as the pulse, accompanied by prepulse bursts categorized as weak (5% of pulse amplitude), medium (10% of pulse amplitude), and high (25% of pulse amplitude) at intensities of 65 dB, 75 dB, and 83 dB respectively. Participants completed one session which contained 42 trials. First, a total of two habituation trials containing white noise bursts of various soft intensities were presented. This allowed for the participant to relax and settle into the movie. After these two trials, eight blocks of five trials were presented. Each block consisted of a pulse only trial and three prepulse-pulse trials. A prepulse always preceded the pulse by 120 ms. Intertrial interval (ITI) was set at random between 10 and 25 seconds. A training session lasted for about 15 minutes.

[0099] Short-term habituation (short-term HAB) paradigms: Participants completed one session which contained 22 stimulus trials. Trials included two practice trials with weak audio pulses which allowed participants to get acclimated to the stimuli. The remaining 20 trials were short-term HAB trials. For HAB trials, pulse trains with six 50 ms white noise audio pulses at 105 dB were used, with the last pulse presented at 4.5 seconds. Two distinct stimulus trams each presented 10 times, in a randomized, but fixed order were used: 1) 10 rhythmic trains were delivered where the pulses were presented at a 1.33 Hz frequency. Pulse number six was delayed by a full cycle so as to create a missing pulse sensation. 2) A more ‘random’ feeling pulse train was delivered 10 times, here the basic rhythm was still 1.33 Hz, but the pulse two to pulse three and pulse four to pulse five interval was shifted to 1 ,25s and 1 s respectively. The ITI was set at random between 20 and 40 seconds. A training session lasted around 15 minutes.

[0100] Analysis of behavioral data: During the presentation of the stimuli, the phone’s forward facing camera and microphone captured the child’s behavioral responses. The generated videos provided a detailed record of the participants’ facial expressions, head position and vocalizations in response to the stimuli. These videos were analyzed for neurobehavioral responses. The following six behavioristic parameters were quantified:

[0101] 1. Screen avoidance: behavior was quantified as the average percentage of frames per trial where the participant’s face was not detected using the Apple Vision framework

[0102] 2. Anteroposterior postural stability: we quantified the variability in face distance from the phone's camera by calculating the size of the bounding box relative to the full field of view (FoV). The relative bounding box size was determined by normalizing the area of the bounding box around the child’s face in each video frame against the total screen area. The dimensions of the bounding box were derived from the differences in bounding coordinates, multiplied by the respective video dimensions. This proportion, rounded to three decimal places, served as a measure of the child's movement extent.

[0103] 3. Head rotations: trained coders observed trial videos to count instances of head rotation defined as the head’s yaw and pitch exceeding 45°.

[0104] 4. Headphone touches: trained coders observed trial videos to count instances where the child used their hands to touch any part of the headphones that were on their head.

[0105] 5. Mouth openings: assessing whether the child's mouth was open, was calculated by analyzing the area of the inner lips in each video frame. The process involves collecting mouth landmark data from the video frames and computing the area enclosed by the inner lip landmarks. The area calculation was performed using a polygon area formula, applicable only if there are three or more points defining the polygon. The formula sums the cross-products of adjacent points, then halves and takes the absolute value of this sum, scaling it by a factor of 20 to enhance readability.

[0106] 6. Vocalizations: trained coders observed trial videos to record timing of vocalizations, as well as their categories. Vocalizations were categorized based on the nature of the sounds rather than their linguistic content (Tenenbaum et al., 2020). Non-syllabic vocalizations included nonverbal vocal sounds not forming part of words. Laughter was separately categorized for its unique expressive and social significance. Lastly, spoken word language encompassed any vocalization resembling structured language, recognized irrespective of the coder’s language proficiency.

[0107] Analysis of neurometric signals (eyelids): Individual eyeblink traces were analyzed in Python 3. 10 with normalization (0 = eye open, 1 = eye fully closed) and filtering as previously described by Boele and colleagues 2023. Utilizing a combination of the smartphone’s forwardfacing camera and microphone enabled us to gather precise positions in regards to eyelid movement. Trials were excluded under several conditions, including the absence of headphones on the child, excessive environmental noise, the child’s face being out of the camera frame, incorrect subject tracking, or issues with the stimuli not being delivered.

[0108] Particular attention was paid to the onset and delivery of each stimulus, with trials exhibiting inconsistent or unpredictable timing being rejected. Trials with extreme outliers in the eyelid signal (signal amplitude < -0.5 normalized eyelid closure, NEC) and trials with spontaneous blinks occurring 150 ms before stimulus onset to 30 ms after stimulus onset were excluded from further analysis. Participants were classified as responsive or unresponsive based on their median startle amplitude (< 0.075 NEC) for pulse-only trials or the first pulse in the habituation paradigms. A total of 2 neurotypical and 3 ASD children were classified unresponsive.

[0109] Acoustic startle response (ASR) analysis: In all participants, the average amplitude of the eyelid startle response to a single 50 ms white noise pulse at 105 dB was studied. The pulse- only trials that were randomly distributed over the PPI experiment were combined with the 20 first pulses delivered during the short-term HAB experiment.

[0110] Long-term habituation (long-term HAB) analysis: The pulse-only trials of the PPI experiment were combined with the first pulses delivered during the short-term HAB experiment. Trials were grouped into five blocks and the average NEC startle amplitude for each block was calculated and compared to investigate the startle amplitude over the full length of a test. It is recognized that this type of paradigm is commonly referred to as habituation, however, as in this disclosure, this is referred to as long-term HAB to avoid confusion with the previously described short-term HAB paradigms. Response detection windows were 300 ms, starting 30 ms after each pulse.

[0111] PPI analysis: The response detection window for eyelid responses to the pulse was set at 30-300 ms after pulse onset. To analyze amplitude reduction, we used the amplitude of the NEC at the mean peak time of significant startle responses calculated over all trials. The variability in startle responses to the PPI stimuli was analyzed by calculating the standard deviation of the amplitude of NEC. Furthermore, the cumulative sum was determined as the median of the cumulative sum of all NEC amplitudes per trial taken across all stimuli types.

[0112] Short-term startle HAB analysis: Response detection windows were 300 ms, starting 30 ms after each pulse (response windows). Short-term habituation was calculated by comparing the NEC startle amplitude across response windows for both the rhythm and random paradigm.

[0113] Anticipatory eye blinks (AEB): In rhythmic pattern trials, 300ms long observation windows were added at 3.63s to capture the 'missing pulse1moment at the center of the window. Similarly, for random pattern trials, 300ms long observation windows were added at 1 ,375s and 3.37s. These windows measure the level in which participants were trying to predict the stimuli in a rhythmic fashion. The cumulative sum of the NEC amplitude was calculated across all response detection and observation windows for the two short-term HAB protocols separately. This was used as a measure to quantify both potential short-term habituation as well as anticipation.

[0114] Statistical analyses and data visualizations used multilevel linear mixed-effects (LME) models. More specifically,

[0115] Statistical analysis and data visualizations were done in Python 3.10 using the packages: Panda’s, Numpy, SciPy, and PyCaret as well as R4.3.1 using the packages: dplyr, emmeans, ggplot2, ImerTest, nlme, tidyr, and tidy verse. Multilevel linear mixed-effects (LME) models were used because they are more robust to violations of normality assumptions, which is often the case in biological data samples. LME models can better accommodate the nested structure of the data (i.e., trial nested within session, session nested within subject, subject nested within group) and prevent data loss by using summary measures. As an added benefit, LME models are better at handling missing data points than repeated measures analysis of variance (ANOVA) models and do not require homoscedasticity as an inherent assumption. In this example LME, session was used as a fixed effect, and subject as a random effect. Goodness- of-fit model comparison was determined by evaluating log likelihood ratio, BIC, and AIC values. The distribution of residuals was inspected visually by plotting the quantiles of standard normal versus standardized residuals (i.e. Q-Q plots). Data were considered as statistically significant if the p-value was less than 0.05. For multiple comparisons, p-values were Bonferonni-Holm adjusted for the number of comparisons made.

[0116] Machine learning

[0117] For model development on the behavioral and neurometric measures, the PyCaret library (Moez, 2020) was used. The analysis focused on evaluating the model's ability to accurately classify cases of ASD. The setup function from PyCaret was configured with the inventor’s dataset, specifying 'has_ASD' as the target variable. Data were z-scored to ensure that the model was not biased by the scale of the different features. To address data imbalances and prevent model bias towards the majority class, synthetic minority over-sampling technique (SMOTE) was applied. The training and testing split was 70% and 30%, respectively, with 10- fold stratified cross-validation. Sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) were calculated using confusion matrices and added to the PyCaret experiment using the add netric function. The comparejnodels function from PyCaret was used to evaluate different models, based on accuracy, area under the curve (AUC), precision, Fl, kappa and Matthew’s correlation coefficient (MCC), sensitivity, specificity, PPV and NPV . For final classification, we used a random forest classifier that was tuned iteratively for optimal specificity by testing combinations of hyperparameters including maximum depth, minimum samples split and leaf, maximum features, bootstrap method and the number of estimators. Model explainability is addressed by calculating SHapley Additive exPlanations (SHAP).

[0118] More specifically, for machine learning for calculating accuracy at individual level: Prior to analysis, data cleaning was performed by replacing infinite and NaN values from the extracted metric of facial landmark time series with Numpy NaN values to ensure data integrity. For the model development, the PyCaret library was used. The setup function from PyCaret was configured with the dataset, specifying 'has_ASD' as the target variable. The session ID was set to '' 123 ’ for reproducibility. The workflow diagram in FIG. 5 illustrates the steps involved in data processing and analysis. Key preprocessing steps included data normalization and handling imbalanced data. The training and testing split was 70% and 30%, respectively, with 10-fold stratified cross-validation. Four custom metrics were defined: sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV). These metrics were calculated using confusion matrices to evaluate model performance in the context of a binary classification problem. The compare models function from PyCaret was used to evaluate different models. The models were compared based on default metrics like accuracy, AUC, precision, Fl, kappa, MCC, including our custom metrics, with results rounded to two decimal places for clarity. A random forest classifier, noted for its robustness and efficiency in handling binary classification tasks, was chosen. The model's out-of-bag score was calculated for an additional performance estimate. Subsequently, the model was tuned using a custom grid search, optimizing for specificity. The tuning process iteratively tested combinations of hyperparameters like maximum depth, minimum samples split and leaf, maximum features, bootstrap method, and the number of estimators. After tuning, the bestperforming model was chosen based on the specified criteria, ensuring an optimal balance between performance and generalizability.

[0119] The analysis focused on evaluating the model's ability to accurately classify cases of ASD. Sensitivity and specificity were key metrics, indicating the model's true positive rate and the ability to correctly identify negative cases, respectively. PPV and NPV provided insights into the model's precision and reliability in predicting positive and negative cases.

[0120] RESULTS The data for girls and boys will be shown combined, as statistically significant diagnosis * sex interactions were not identified after correcting for multiple comparisons. All mean values reported below are ± standard deviation (SD) and p-values are Bonferonni-Holm adjusted.

[0121] Screen avoidance: Children with ASD avoided watching the smartphone's screen significantly more than neurotypical children (F 1,278=33.13, pO.OOOl; see FIGS. 3A, 6). On average, children with ASD displayed this behavior in 17.73% (±19.62) of trials, while neurotypical children faced away in 5.27% (±11.40) of trials.

[0122] Anteroposterior postural stability: Children with ASD exhibited significantly more anteroposterior postural movements during the experiments compared to their neurotypical counterparts (Fi, 278=22.48, p<0.0001; see FIGS. 3B, 6). Children with ASD demonstrated a mean standard deviation of the facial bounding box size of 9.96 (±12.64), whereas neurotypical children exhibited a value of 3.79 (±2.66).

[0123] Head rotations: Children with ASD demonstrated significantly more head rotations compared to neurotypical children (Fi, 278=23.93, pO.OOOl; see FIGS. 3C, 6). Children with ASD exhibited head rotations in 14.91% (±14.94) of trials while neurotypical children displayed head rotations in 6.74% (±9.44) of trials.

[0124] Headphone touches: Children with ASD displayed significantly more headphone touches during the experiments (F 1,278=31.04, pO.OOOl; see FIGS. 3D, 6). Children with ASD displayed headphone touches in 10.60% (±15.53) of trials, contrasting with neurotypical children who exhibited this behavior in 1.71% (±3.25) of trials.

[0125] Mouth opening: Children with ASD displayed both an increased frequency and larger mouth openings compared to neurotypical children (Fi, 278=55.63, pO.OOOl; see FIGS. 3E, 3G, 6). The mean mouth surface area for children with ASD was 11.53 (±10.29) versus a mean of 3.31 (±4.72) in neurotypical children. In addition, mouth opening was elicited with auditory stimuli in children with ASD, a response not observed in neurotypical controls (see FIG. 3G).

[0126] Vocalizations: Children with ASD demonstrated a significantly higher percentage of both non-syllabic (F 1,278=47.58, pO.OOOl; see FIGS. 3F, 3G, 6) and laughter (Fi, 278=8.93, p=0.0061; see FIGS. 3G, 6) vocalizations during the experiments. Notably, both types of vocalizations occurred spontaneously throughout the 15-minute experiment. Furthermore, similar to mouth opening, these vocalizations were elicited with audit ory stimuli in children with ASD, a response not observed in neurotypical controls (see FIG. 3G). Non-syllabic vocalizations were observed in 12.00% (±16.97) of trials in ASD children, in contrast to 0.10% (±0.49) of trials in neurotypical children (see FIG. 3F). There was a significant effect of sex on non-syllabic vocalizations (F 1,276= 10. 18, p=0.012), with a higher percentage observed in girls than boys; however, no significant sex * diagnosis interaction was found (F 1,276=6. 13, p=0.11). Regarding laughter, children with ASD exhibited this category of vocalization in 1.56% (±5.05) of trials, whereas neurotypical children displayed it in 0.03% (±0.25) of trials.

[0127] Acoustic startle response (ASR): No significant difference in the amplitude of the eyelid ASR was detected between the ASD and neurotypical children (Fi, 268=0.40, p=0.53; see FIGS. 7A, 7B). The mean eyelid startle amplitude for ASD children was 0.23 (±0.21), while in neurotypical children it was 0.19 (±0.21). There was a significant effect of sex (Fi, 266=7.25, p=0.0075; see FIGS. 7A, with a marginally significantly higher startle amplitude in girls with ASD compared to boys with ASD 0266=2.63, p=0.036), however, no significant sex * diagnosis interaction was found (Fi, 266=0.75, p=0.39).

[0128] Long-term habituation (long-term HAB): ASD and neurotypical children showed comparable levels of habituation of the ASR (see FIGS. 7C, 7D) over the course of the 15- minute test. In children with ASD there was a decrease in startle amplitude from 0.26 (±0.24) in trials 1-2 to 0.18 (±0.18) in trials 9-10 (see FIGS. 7C, 7D). Similarly, in neurotypical children, the mean ASR decreased from 0.26 (±0,25) in trials 1-2 to 0. 17 (±0. 19) in trials 9-10. While the main effect of trial number was significant (F 4, 4231=13.57, p<0.0001), the diagnosis (Fi, 268=1.99, p=0.16) and trial number * diagnosis interaction effects (F4, 4231=0.60, p=0.66) were not significant.

[0129] Prepulse inhibition (PPI): Children with ASD demonstrated reduced levels of PPI and increased variability in their startle eyelid responses (see FIGS. 7E-7I). In neurotypical children, a statistically significant reduction in the amplitude of the eyelid startle response was found as the prepulse intensity increased, whereby the normalized eyelid closure (NEC) progressively decreased from 0.35 (±0.24) in trials without a prepulse to 0.25 (±0.20) in trials with the maximum prepulse intensity at 25% of the pulse intensity 01626=4 45, p<0 0001). In contrast, children diagnosed with ASD did not exhibit a corresponding amplitude reduction. The average NEC in pulse-only trials was 0.37 (±0.22) and in trials with the highest prepulse intensity it was 0.35 (±0.24) 02083=1.44, p=0.45). A statistically significant effect was found of diagnosis for the trials with the prepulse intensity at 25% (F 1,152=7.60, p=0.020). Interestingly, children with ASD exhibited significantly increased variability in response amplitudes for all prepulse intensities (figure 7G-I). Analyzing the cumulative summation of eyelid amplitudes across the entire trace, regardless of trial type, underscored a consistent disparity between neurotypical and ASD groups (Fi, 156=11.92, p=0.00072). Short-term habituation (short-term HAB): In both the rhythmic and random stimulus patterns, no decrease (i.e., habituation) was observed in the mean amplitude of eyelid startle responses over the course of the six startle pulses. This lack of habituation was consistent across both children with ASD and neurotypical children (FIGS. 8A, 8B). Notably, reflexive startle amplitudes tended to increase slightly in the ASD group after the first pulse, with a statistically significant startle amplitude increase at pulse 3 in the random stimulus pattern (pulse 1 vs. pulse 3: t56o= -2.78, p=0.029).

[0130] Anticipatory eye blinks (AEB): For the rhythmic pulse train, no AEBs were found in neurotypical children (NEC of 0.05 ±0.09), whereas ASD children showed a small, but statistically significant increase (NEC of 0.09 ±0.11; Fi,i2o=4.81 , p=0.030) (see FIGS. 8A, 8E).

[0131] For the random pulse train, strong, statistically significant AEBs were found (Fi, ii2=7.33, p=0.0078 for OW-1 and Fi,ii2=l 1.78, p=0.0017 for OW-2) in the two predefined observational windows for the ASD children (NEC of 0.13 ±0.15 for OW-1 and 0.12 ±0.12 for OW-2), but not for neurotypical children (NEC of 0.06 ±0.09 for OW-1 and 0.04 ±0.09 for OW-2), see FIGS. 8B, 8F. To quantify the short-term HAB in conjunction with the AEB, the cumulative summation of the mean eyelid amplitude in the two short-term HAB protocols were examined. The cumulative sum for the rhythmic protocol was not significantly different (Fi, 120=3.77, p=0.055, see FIG. 8C) between children with ASD (cumulative NEC=54.02 ±45.13 SD) and neurotypical children (cumulative NEC=39.03 ±34.86). For the random stimulus pattern, the mean cumulative sum of the eyelid amplitude was significantly higher (Fi, 112=8.61, p=0.0041, see FIG. 8D) in children with ASD, (cumulative NEC=65.32 ±47.36), compared to typically developing children (cumulative NEC=40.81 ±35.30).

[0132] Binary classification results: Neurobehavioral outcomes were used as features in a machine learning binary classification. Feature strength was evaluated using Pearson's correlation coefficient, revealing strong correlations with the target ASD for both behavioral and neurometric features, without significant multicollinearity (see FIGS. 9A and FIG. 5).

[0133] Diagnostic accuracy was further assessed by comparing different machine learning models. Note that all models performed reasonably well, except for the dummy classifier, which was added for baseline comparison. In the example dataset, the tree-based algorithms, specifically the random forest, gradient boosting, and extra trees classifiers, showed the best performance. The random forest classifier was further evaluated based on a stratified 10-fold cross-validation approach and a holdout set evaluation.

[0134] For the holdout set, the classifier achieved an accuracy of 0.84, with a precision of 0.90 and a recall of 0.84. The Fl score was noted at 0.87. Sensitivity and specificity were both reported at 0.84 and 0.83, respectively. The PPV was 0.90, and the NPV was 0.74, with an AUC of 0.83. The model's kappa score was 0.65, and the MCC was also 0.65. The OOB score for the model was 0.87. In terms of confusion matrix metrics, the model identified 46 true positives (TP), 25 true negatives (TN), 5 false positives (FP), and 9 false negatives (FN). See FIG. 9B. Note that the results from the cross-validation process revealed a similar performance.

[0135] Model explainability was addressed by calculating SHAP values for each entry in the holdout set. Features like 'non-syllabic vocalizations' and 'screen avoidance', which exhibited higher SHAP values, were indicative of greater significance in the model's decision-making process (see FIG. 9C). To further investigate the diagnostic utility of each individual feature, a random forest classifier was trained for each measurement. See FIG. 10.

[0136] Features derived from neurometric tests (PPI and AEB) proved effective in accurately identifying positive ASD cases, demonstrating high sensitivity. Conversely, more general behaviors, such as non-syllabic vocalizations and headphone touches, were particularly adept at confirming the negative cases, indicated by their high specificity. See FIG. 10.

[0137] In this example, sensorimotor anomalies linked to ASD were measured to assess the diagnostic accuracy of smartphone-based neurobehavioral evaluations. The findings revealed increased levels of spontaneous and stimulus-evoked postural, head, facial, and vocal responses in children with ASD. Additionally, the ASD group exhibited diminished PPI levels and heightened anticipatory eye blinks. By employing a random forest machine learning algorithm for binary classification based on the combined neurobehavioral outcomes, a diagnostic accuracy of 84% was obtained. Comparable results were obtained in the cross-validation set and the holdout set in terms of sensitivity (cross-validation 85% vs. holdout 84%) and specificity (cross-validation 84% vs. holdout 83%), indicating that the model generalizes well to unseen data and maintains a balanced performance between correctly identifying positive and negative cases across different datasets. This consistency enhances confidence in the model's reliability and robustness in real-world applications. Model explainability was addressed by calculating SHAP values for each parameter in the holdout set (figure 5, table S13, 14). In addition, the diagnostic utility of individual features was investigated by training a random forest classifier for each outcome (see FIG. 10). It was found that some behavioral features, such as non-syllabic vocalizations and headphone touches, were indicative of greater significance in the model's decision-making process, but often lacked sensitivity. On the other hand, features derived from PPI and AEB were mostly effective in accurately identifying the positive ASD cases, demonstrating strong sensitivity (see FIG. 10). The low multi collinearity of these combined neurobehavioral features synergistically enhances the robustness and diagnostic accuracy of the model for ASD. The platform’s rapid evaluation speed, low burden on children and their caregivers, and high predictive value collectively position it as a promising candidate in the search for novel obj ective and accessible tools that can help alleviate the waitlist crisis in autism.

[0138] Several of the spontaneous and evoked behavioral parameters acquired with the disclosed digital platform showed significant differences among the diagnostic groups. Particularly, non-syllabic vocalizations were more common in children with ASD, aligning with prior research. Likewise, postural movements and mouth openings were increased in children with ASD, as well as head rotations, headphone touches, and screen avoidance. These results are compatible with other studies using video presentations. Similar to studies done in an academic laboratory setting, diminished levels of PPI were found in terms of eyelid closure amplitude. In addition, it was found that children with ASD had increased variability in their response amplitudes. This observation highlights that hypo- or hyperresponsiveness to sensory stimuli is an oversimplification when it comes to ASD sensory anomalies.

[0139] By using both rhythmic and random stimulus trains, the level of short-term habituation was determined. Neither neurotypical nor children with ASD showed decrements in their responses following repeated stimulation with the same stimulus. This coincides with another study that also found weak or even no habituation in children. Unlike neurotypical children, children with ASD showed a slight increase in their startle amplitudes after the first pulse, and may reflect stronger sensitization in children with ASD.

[0140] Resistance to change upon presentation of any type of sensory stimulation is an important symptom of ASD, both in humans and animal models. It is speculated that this resistance to change is reflected by the higher levels of anticipatory eye blinks in the children with ASD, since these blinks were most obvious at times during the trials where there was a clear omission of an auditory stimulus. These anticipatory responses may be indicative of a hyper-responsiveness to the “missing” stimulus.

[0141] Various modifications may be made to the systems, methods, apparatus, mechanisms, techniques and portions thereof described herein with respect to the various figures, such modifications being contemplated as being within the scope of the invention. For example, while a specific order of steps or arrangement of functional elements is presented in the various embodiments described herein, various other orders / arrangements of steps or functional elements may be utilized within the context of the various embodiments. Further, while modifications to embodiments may be discussed individually, various embodiments may use multiple modifications contemporaneously or in sequence, compound modifications and the like.

[0142] Although various embodiments which incorporate the teachings of the present invention have been shown and described in detail herein, those skilled in the art can readily devise many other varied embodiments that still incorporate these teachings. Thus, while the foregoing is directed to various embodiments of the present invention, other and further embodiments of the invention may be devised without departing from the basic scope thereof. As such, the appropriate scope of the invention is to be determined according to the claims.

Claims

What is claimed is:

1. A system for detecting a neurodevelopmental condition, comprising: one or more processing units configured to, collectively: cause an auditory stimulus to be experienced by a user; receive audio and video of the user for a period of time; make an auditory assessment by determining if the audio includes syllabic vocalizations, non-syllabic vocalizations, or both, in response to an auditory stimulus; make a visual assessment by determining at least two spontaneous unstructured movements in response to the auditory stimuli, the at least two spontaneous unstructured movements including a degree of openness of at least one eye and at least one additional spontaneous unstructured movement; and determine a neurodevelopmental condition is present based on the auditory assessment and / or the visual assessment.

2. The system of claim 1, wherein the at least one additional spontaneous unstructured movement includes: a degree of openness of a mouth; a degree of head movement; an arm or hand movement; or a combination thereof.

3. The system of claim 2, wherein the arm or hand movement is determined to be the user reaching for an ear or the user covering an ear.

4. The system of any one of claims 1-3, wherein the auditory stimulus is white noise.

5. The system of any one of claims 1-3, wherein the auditory stimuli is one or more discrete frequencies of sound.

6. The system of any one of claims 1-5, wherein the period of time is no more than 5 seconds.

7. The system of any one of claims 1-6, wherein the period of time is no more than 2 seconds.

8. The system of any one of claims 1-7, wherein the neurodevelopmental condition is autism spectrum disorder (ASD).

9. The system of any one of claims 1-7, wherein the neurodevelopmental condition is attention-deficit / hyperactivity disorder (ADHD).

10. The system of any one of claims 1-9, wherein the auditory assessment is informed by research findings on vocalization patterns in individuals with the neurodevelopmental condition.

11. The system of any one of claims 1-10, wherein mouth state data is used in conjunction with at least one other behavioral indicator to assess the neurodevelopmental condition.

12. The sy stem of any one of claims 1-11, wherein the visual assessment is informed by research findings on head movements in individuals with the neurodevelopmental condition.

13. The sy stem of any one of claims 1-12, wherein the neurodevelopmental condition is determined based solely on the auditory assessment.

14. The system of any one of claims 1-12, wherein the neurodevelopmental condition is determined based solely on the visual assessment.

15. The system of any one of claims 1-14, wherein the audio and video of the user received by the one or more processing units includes a period of time prior to the user experiencing the auditory stimuli.

16. The system of any one of claims 1-15, wherein the audio and video of the user received by the one or more processing units includes the user experiencing the auditory stimuli and a period of time after the user experienced the auditory stimuli.

17. The system of any one of claims 1-16, further comprising:a first display; a camera; a speaker; a memory; and where the one or more processing units are operably coupled to the display, camera, speaker, and memory.

18. The system of claim 17, wherein at least one of the one or more processing units are configured as a desktop computer, laptop computer, mobile phone, or tablet.

19. The system of claim 17 of 18, wherein the one or more processing units includes a local processor and a remote processor, where the remote processor is configured to perform the auditory and visual assessment steps.

20. The system of any one of claims 17-19, wherein the display and camera are operably coupled to a headset, the headset being operably coupled to the one or more processing units.

21. The system according to claim 20, wherein the speakers are coupled to the headset.

22. The system according to any one of claims 17-21, wherein the speakers are configured as headphones.

23. A non-transitory computer-readable storage device containing instructions that, when executed by one or more processing units, causes the one or more processing units to, collectively, perform a method, the method comprising: exposing an individual to auditory stimulus; recording any vocalizations, head movements, and / or mouth movements of the individual; identifying one or more patterns by analyzing the vocalizations, head movements, and / or mouth movements; determining if the one or more patterns are associated with a neurodevelopmental condition; and providing an assessment.

24. The method of claim 23, wherein the assessment is provided to a clinician or therapist.

25. A method for measuring unstructured behavioral responses utilizing a system of any one of claims 1-22, comprising: receiving an assessment, wherein the assessment has been generated for an individual following a computer-based assessment system utilizing a system of any one of claims 1-22, wherein the computer-based assessment system is configured to: expose an individual to auditory stimulus; record any vocalizations, head movements, and / or mouth movements of the individual; identify one or more patterns by analyzing the vocalizations, head movements, and / or mouth movements; determine if the one or more patterns are associated with a neurodevelopmental condition; and provide an assessment; and based on the assessment, instituting or adjusting therapy for the individual responsive to the neurodevelopmental condition.

26. The method of claims 25, further comprising instructing the individual to utilize the computer-based assessment system.

Images