Prediction of future autism diagnostics and intervention responses using neural data and machine learning

Abstract

A machine learning model based on neural data (e.g., electroencephalogram (EEG) data, magnetoencephalogram (MEG), and / or magnetic resonance imaging (MRI) data) may be used to predict the effectiveness of future diagnoses and / or interventions (e.g., early interventions) of autistic (autistic) spectrum disorders (ASD) in an individual child. The training of the model may be based on experimental data obtained in a child who previously accepts an intervention.

Description

Cross - Reference to Related Applications

[0001] This application claims the benefit of U.S. Provisional Application No. 63 / 594,377, filed on October 30, 2023, the disclosure of which is incorporated herein by reference. Background of the Invention

[0002] This disclosure generally relates to predicting the outcomes of children receiving interventions related to autism and, in particular, to using neural data to predict future diagnoses of autism and responses to interventions.

[0003] Autism spectrum disorder (ASD) refers to a developmental disorder caused by differences in the brain. Globally, the prevalence of ASD has increased sharply in the past decade. For example, in China, the prevalence increased from 0.39% in 2000 to 0.7% in 2016. In the United States, the Centers for Disease Control and Prevention (CDC) estimates the prevalence in the United States in 2021 to be 2.3%. ASD varies widely in severity and presentation.

[0004] ASD is typically diagnosed based on behavioral symptoms such as problems related to social communication and social interaction, restrictive or repetitive behaviors, and delays in various developmental schedules. Although it is believed that the brain differences causing ASD are present at birth, a formal diagnosis of ASD is currently only possible when the child is old enough for the behavioral symptoms to become apparent. Generally, ASD cannot be diagnosed in children younger than about 4 years old.

[0005] Early intervention (EI), including behavioral intervention, has shown promise in reducing the severity of the impact of ASD. For example, in one form of behavioral intervention, parents are coached to implement communication strategies at home to enhance the child's social communication. Clinicians and researchers agree that waiting until age four to start EI is not optimal, and in some cases, EI treatment can be preemptive (before an ASD diagnosis) with good outcomes.

[0006] However, EI is not always necessary or effective. When examining the improvement of individual children in detail, a large variability in the response to EI can be observed. While some children improve a lot, other children may not improve or may even regress, and improvement or regression cannot be predicted by the child's age or gender. In addition, there are different forms of EI, and different children are expected to benefit from different interventions. Summary of the Invention

[0007] Certain embodiments of the present invention relate to techniques for predicting the effectiveness of future diagnosis and / or intervention (e.g., early intervention) for ASD in individual children. The prediction can be based on neural data, such as electroencephalogram (EEG) data, magnetoencephalogram (MEG), and / or magnetic resonance imaging (MRI) data. The neural data can include structural and / or functional data. Quantitative data extracted from the neural data can be provided to a machine learning model, such as a classifier, that has been trained to predict the effect of a particular intervention. The training can be based on experimental data obtained from children who have previously undergone an intervention; the data can include neural data obtained before the intervention, a baseline score of an assessment of ASD-related characteristics of the children implemented before the intervention, and a second score of the assessment implemented after the intervention.

[0008] The following detailed description and the accompanying drawings will provide a better understanding of the features and advantages of the claimed invention. Brief Description of the Drawings

[0009] Figure 1 A flowchart showing a process for training a machine learning model to predict the future ASD diagnosis and / or the effectiveness of a proposed intervention for an individual child, according to an embodiment of the present invention.

[0010] Figure 2 A graph showing the receptive communication scores of an individual child before and after an intervention aimed at enhancing the child's social communication.

[0011] Figure 3 A flowchart of a process for predicting EI outcomes for a child, according to an embodiment of the present invention.

[0012] Figure 4 Summarizes the results obtained from training various machine learning models according to some embodiments.

[0013] Figure 5 A circular bar chart showing the contribution weights of neural and non-neural features in predicting the response to an intervention using a machine learning model, according to some embodiments.

[0014] Figure 6 A correlation graph showing the correlation between neural EEG data at a first time and ADOS-2 scores that can be obtained at a later date, according to some embodiments.

[0015] Figures 7A to 7C Shows the correlation between the functional connectivity of brain structures and scores regarding ASD assessment. Figure 7A Shows a brain structure. Figure 7B A scatter plot showing the ADOS-2 social impact scores of 15 children versus the FC of the amygdala and inferior frontal cortex. Figure 7CA scatter plot showing the ADOS-2 restricted / repetitive behavior scores of 15 children versus the FC of the amygdala and posterior cingulate cortex is presented. Detailed Description

[0016] For purposes of illustration and description, the following description of exemplary embodiments of the invention is given. It is not intended to be exhaustive of the claimed invention or to limit the claimed invention to the precise forms described, and those skilled in the art will understand that many modifications and variations are possible. The embodiments have been chosen and described in order to best explain the principles of the invention and its practical applications, so that those skilled in the art can best make and use the invention in various embodiments and the various modifications suitable for the specific uses contemplated. Overview

[0017] Figure 1 A flowchart of process 100 for training a machine learning model to predict the future ASD diagnosis and / or the effectiveness of a proposed intervention for an individual child, according to an embodiment of the invention, is shown. Process 100 can be implemented using a suitably programmed computer system.

[0018] At block 102, a training data set is prepared. The training data set can include information obtained from children who have been diagnosed with ASD and have undergone an intervention (e.g., early intervention, or "EI") aimed at reducing the severity of ASD. (It should be understood that the ASD diagnosis can be made before or after the intervention.) The information can include neural data obtained by scanning or monitoring the brain of the child before the intervention. Examples of neural data include electroencephalogram (EEG) data, magnetoencephalogram (MEG), and / or magnetic resonance imaging (MRI) data. The neural data can include structural and / or functional data. For example, EEG data, MEG data, and / or MRI data can be recorded while the child is at rest and while the child is listening to speech. The neural data can be preprocessed, for example, to reduce noise, resample images, select target brain regions, (e.g., by applying spectral analysis techniques to EEG data) reduce the size of the data set, etc. In some embodiments, the neural data can be obtained at some point in time before the child's formal ASD diagnosis (e.g., between birth and about 4 years of age).

[0019] In some embodiments, additional data about the child can also be included. For example, "pediatric" data can include gender at birth, age at assessment, level of family education, family income, and other demographic data. "Psychological" data can include, for example, one or more developmental assessment scores determined for the child using an assessment tool (e.g., Autism Diagnostic Observation Schedule, Second Edition (ADOS-2)).

[0020] Information included in the training dataset can also include information quantifying a child's response to an intervention. For example, behavioral scores of a child before and after an intervention can be determined using standard instruments that measure the specific capabilities of the intervention aimed at improvement. In one example, the intervention requires parental guidance to improve social communication skills. Standard assessment tools (such as the Vineland Adaptive Behavior Scale) can be used to measure communication skills. Assessments can be administered before the intervention and after a period of time (e.g., 6 to 8 months) of the intervention. The period of the intervention can be selected based on the specific intervention and should be long enough to elicit an expected response.

[0021] As an example, Figure 2 A graph showing receptive communication assessment scores (determined using the Vineland Adaptive Behavior Scale) of 58 individual children before and after an intervention, where parents were coached to implement communication strategies at home during the intervention to enhance the children's social communication. The x-axis indicates the age of the child (in months) at the time of determining the score, and the y-axis indicates the score; higher scores correspond to higher capabilities. The scores of the same child before and after the intervention are connected by a line. The response of the child can be quantified as the difference in scores. As Figure 2 shown, there is significant variability in the response to the intervention, with some showing significant improvement, some showing small improvement, and a few even regressing. This variability has no significant correlation with age, gender, or the initial (pre-intervention) assessment scores.

[0022] Referring again to Figure 1 , at block 104, quantitative data can be extracted from the neural data. For example, an EEG can include hundreds or thousands of data samples. Spectral analysis and / or other techniques (such as comparing datasets obtained under different stimulus conditions or in the absence of a stimulus condition) can be used to quantify the characteristics of the dataset. For an MRI, various techniques can be applied to select the voxels corresponding to the target region and use the reduced-dimensional dataset to characterize the features of such regions.

[0023] At block 106, an automatic classification algorithm (also referred to as a "classifier" or "machine learning model") can be trained using the training dataset, which includes the quantitative data extracted from the neural data. Suitable algorithms include classification algorithms for machine learning, such as support vector machines (SVMs), ranked SVMs (RankSVMs), or random forests.

[0024] SVM is a classification technique in machine learning that takes as input a feature vector and a binary classification in a space of any dimension, and maps the feature vector to a point in a classification space such that a hyperplane (referred to as the "margin") in the classification space separates the points corresponding to the (binary) classification of the respective feature vectors. In most SVM implementations, the margin can be a "soft margin" that allows for less than 100% accuracy in classification. In this context, the feature vector can be voxel data generated for a given child, and based on the magnitude of the difference between the pre-intervention and post-intervention assessment scores for the child, the binary classification can be "low improvement" or "high improvement".

[0025] RankSVM is a classification technique in machine learning whose goal is to construct an ordered model that can be used to classify unseen data according to its relevance or importance. RankSVM can be used to form a ranking model by minimizing a pairwise loss based on a regularized margin. RankSVM uses SVM to calculate a weight vector that maximizes the difference between data pairs in the ranking. In principle, RankSVM needs to investigate each data pair as a potential candidate for a support vector, and the number of data pairs is the square of the training set size. In practice, this can lead to reduced computational efficiency for large training sets and / or large feature vectors. Therefore, optimizations can be employed to improve computational efficiency; specific examples are described below.

[0026] Random forest is a machine learning prediction or classification technique based on an ensemble of decision trees. Each decision tree can be trained to make predictions based on a random subset of features, and different subsets of the training data can be used to train different decision trees. The final prediction can be made by taking the average or mode of the predictions of the decision trees.

[0027] The training of a machine learning model involves automated processing to determine or "learn" the optimal values of the internal parameters of the model, such as the weights of each node or the coefficients of parametric functions (e.g., curve fitting functions or transformation functions). Standard methods of training involve repeatedly processing data samples through the model and adjusting the model's parameters with the aim of minimizing a loss function that characterizes the difference between the model output for a given input and an expected outcome, which is determined from a source other than the model. In the embodiments described herein, the expected outcome for each input training data point can be based on post-intervention assessment. The loss function can be selected, in part, based on the particular model, and various techniques can be used to optimize the loss function. Training typically occurs over multiple "epochs", where each epoch corresponds to a complete traversal of the training sample set. Adjustment of the model's parameters (e.g., weights or coefficients) can occur multiple times during an epoch; for example, the training data can be divided into "batches" or "mini-batches", and weight adjustment can occur after each batch or mini-batch. Aspects of machine learning models and training relevant to understanding the present disclosure are described herein; any other aspects can be modified as needed.

[0028] Referring again to Figure 1 , once the machine learning model has been trained and validated, the trained machine learning model can be stored (block 106) for future application to children who are candidates for intervention.

[0029] Figure 3 The use of the trained machine learning model is shown, Figure 3 is a flow chart of a process 300 for predicting the outcome of a child intervention according to an embodiment of the present invention. At block 302, for example, a baseline assessment score for the child is determined by implementing the same assessment used for the children included in the training data. At block 304, neural data for the child is obtained. The same techniques and manner used to obtain the neural data for the children in the training dataset are used. Other data for the child can also be obtained, including pediatric data and / or psychological data (e.g., scores on other developmental assessment tests). At block 306, quantitative data is extracted from the neural data. The same techniques and manner used to extract quantitative data from the neural data for the children in the training dataset are used. At block 308, the trained machine learning model is used to analyze the data. The machine learning model applies the trained algorithm to output a classification result, e.g., predicting the degree of symptom improvement based on the intervention measures received by the children in the training dataset. Depending on the particular machine learning model, the predicted degree of improvement can correspond to a binary "low improvement" or "high improvement" classification or to a quantified degree of improvement. At block 306, the predicted degree of improvement and the child's baseline ASD score can be used to generate a prediction result. This prediction result can be provided to a clinician for treatment planning.

[0030] Those skilled in the art will appreciate that a process similar to process 300 can also be used to validate a model during training process 100. Training samples can be input into the model being validated to obtain "predicted" results. In this case, the actual results are also known, and comparing the "predicted" results with the actual results provides an indication of the accuracy of the model.

[0031] In some embodiments, the model can be self-updating. For example, once constructed, the model can be used to evaluate whether new (previously unobserved) children are suitable candidates for intervention. When children who are evaluated using the model receive the intervention and their outcomes have been determined, the data of these children can be added to the dataset used for training, and the model can be updated in real time (e.g., by repeating processes 100 and 200 using the enlarged dataset). Then, the updated model can be used to predict the outcomes of other children.

[0032] In some embodiments, a machine learning model can be trained to predict future ASD diagnoses without any intervention other than allowing time to pass. For example, training data can be obtained from children who have not received an intervention. The training data can include neural data obtained at a first time, a baseline assessment of ASD-related characteristics at or near the first time, and a follow-up assessment of the same ASD-related characteristics at a second time, which can be, for example, several months or a year or more after the first time. Between the first time and the second time, the children can be allowed to develop in their natural environment without receiving any specific intervention aimed at alleviating ASD symptoms or behaviors. The machine learning model trained using this data can predict or foretell future ASD diagnoses without active intervention. In some embodiments, such predictions can be used as a "control" and compared with predictions of active intervention. Examples

[0033] To further illustrate these processes, specific examples of a classification model that has been trained to predict the outcomes of EI in an experimental context will now be described. These examples are based on data from 58 children who received EI in the form of parent coaching as described above. At the time of participating in the intervention, the children's ages were between 25 and 54 months. They had been evaluated by qualified medical professionals and were assessed as having autism or a high likelihood of autism and requiring close monitoring. Each child was also evaluated using the ADSOS-2 and scored within the autism or autism spectrum range, and the Vineland Adaptive Behavior Scales were used to measure the receptive communication abilities of the children before and after receiving the intervention.

[0034] Several machine learning models using the random forest algorithm were trained to predict improvements in receptive communication ability. Different models received different combinations of data. The first model, called "Pedi", received only pediatric data, including sex at birth, age at assessment, level of family education, and family income. The second model, called "Psy", received the child's developmental assessment scores, including ADOS-2 and initial Vineland scores. The third model, called "Neuro", received EEG data including the child's language neural coding ("frequency following response" data) and resting state ("rest") EEG data. Other models received combinations of these data types, including "Pedi+Psy", "Pedi+Neuro", and "Pedi+Psy+Neuro". Each model was trained multiple times using different random subsets of the training data and validated using the remaining training data.

[0035] Figure 4 Summarizes the results obtained from training and validating various machine learning models. Different models are listed along the x-axis; the y-axis depicts the distribution of the area under the receiver curve (AUC) for each model. The figure also shows the random permutations used for comparison. The asterisk indicates the median AUC. As Figure 4 shown, models that include neural data outperform models that do not include neural data. An AUC of 0.8 can be considered a threshold for clinical significance, and only models that include neural data can reach that threshold.

[0036] Figure 5 A circular bar chart was used to show the contribution weights of the neural and non-neural features of different machine learning model prediction intervention responses. Each sector within the circle corresponds to a different feature input into the machine learning model. Neural features are labeled in red (names starting with "FFR" or "Rest"), while non-neural features are labeled in white. The size of the bar corresponds to the weight of each feature. It can be seen that neural features always have the highest weight. Computer implementation

[0037] The data analysis and computational operations described herein can be implemented in a computer system of conventional configuration, such as a desktop computer, laptop computer, tablet computer, mobile device (e.g., smartphone), etc. Such systems can include one or more processors to execute program code (e.g., a general-purpose microprocessor that can serve as a central processing unit (CPU) and / or a special-purpose processor, such as a graphics processing unit (GPU), which can provide enhanced parallel processing capabilities); a memory and other storage devices for storing program code and data; user input devices (e.g., a keyboard, a pointing device such as a mouse or touchpad, a microphone); user output devices (e.g., a display device, speakers, a printer); combined input / output devices (e.g., a touchscreen display); signal input / output ports; a network communication interface (e.g., a wired network interface such as an Ethernet interface and / or a wireless network communication interface such as Wi-Fi); etc.

[0038] A computer program incorporating features of the present invention that can be implemented using program code can be encoded and stored on various computer-readable storage media; suitable media include magnetic disks or tapes, optical storage media (e.g., a compact disc (CD) or a digital versatile disc (DVD)), flash memory, and other non-transitory media. (It should be understood that the “storage” of data is different from the propagation of data using a transitory medium such as a carrier wave.) A computer-readable medium encoded with program code can include an internal storage medium of a compatible electronic device, and / or an external storage medium readable by an electronic device with the executable code. In some cases, the program code can be provided to the electronic device via the Internet or other transmission paths.

[0039] In an alternative embodiment, a special-purpose processor can be used to perform some or all of the operations described herein. Such a processor can be optimized, for example, for performing calculations to train a random forest model or other machine learning models, and can be incorporated into other conventionally designed computer systems or other computer systems. Additional Embodiments

[0040] In some embodiments, a machine learning model of the above type can be trained to predict future ASD diagnoses. For example, it has been observed that neuro-EEG data at a first time is correlated with the scores of the ADOS-2 obtained at a later date. Figure 6Shows the effect of a specific feature (FFR_PCA2) extracted from EEG data. As shown in a sample of 15 children, FFR_PCA2 was correlated with the ADOS-2 score obtained 12 months later (r = 0.47, p = 0.037 [one-tailed]). It is expected that other EEG features may also be correlated with future ADOS-2 scores, and training machine learning models using multiple EEG features and / or other neural data can enable predictive diagnosis of autism based on neural data (e.g., before children are old enough for routine diagnosis).

[0041] In some embodiments, in addition to or instead of EEG data, MRI data can also be used to predict the severity of autism. For example, it has been observed that functional connectivity measurements derived from functional MRI data are correlated with assessment scores related to ASD. Figure 7A Shows brain structures including the amygdala 702, inferior frontal cortex 704, and posterior cingulate cortex 706. Functional connectivity (FC) is indicated by the blue arrows between the structures. Figure 7B Shows a graph of the ADOS-2 social impact scores of 15 children versus the FC of the amygdala and inferior frontal cortex. A correlation was observed (r = -0.59, p = 0.020). Similarly, Figure 7C Shows a graph of the ADOS-2 restricted / repetitive behavior scores of 15 children versus the FC of the amygdala and posterior cingulate cortex. A correlation was observed (r = -0.63, p = -0.013). Thus, fMRI data is expected to be an effective input metric for predicting the effect of EI and / or predicting future ASD diagnosis.

[0042] Although the present invention has been described with reference to specific embodiments, those skilled in the art will understand that changes and modifications can be made. For example, although the examples described above focus on early intervention to improve communication skills, particularly receptive communication skills, other interventions can target other symptoms of ASD. Using the techniques described herein and appropriate assessments to measure the effect of the intervention on an individual basis, the effectiveness of other interventions can be predicted.

[0043] Neural data can include data representing the structure and / or function of any part of a child's central nervous system, including the brain. Structural data can be obtained, for example, using MRI. Functional data can be collected using EEG, MEG, functional near-infrared spectroscopy (fNIRS), functional magnetic resonance imaging (fMRI), or other means. Neural data collected using different means can be combined or used separately. Neural data can include data representing the entire brain, or data for specific regions or structures within the brain, or data for other parts of the central nervous system. In some embodiments, some data can be acquired while the child is sedated or in natural sleep, and some data can be acquired while the child is exposed to stimuli (such as language or visual stimuli).

[0044] A variety of techniques can be used to extract quantitative data from neural data. For example, spectral analysis techniques or the like can be used to extract quantitative data from EEG data. For imaging data, such as MRI or fMRI data, image registration and masking techniques can be used to select target regions, and various image analysis techniques can be applied to extract features from the images to reduce the size of the feature vectors input to a machine learning model. Data obtained under different conditions (e.g., at rest versus when exposed to a stimulus or when exposed to two or more different stimuli) can be compared to determine differences, and these differences can be used to define quantitative data.

[0045] Classification of children according to degree of improvement can be based on any measurement of communication skills and / or behavior, including but not limited to the Vineland assessment, ADOS-2, or other tests described above. For a given child, the baseline (prior to EI) assessment can be used as a reference point for predicting improvement, thus allowing a comprehensive assessment of possible outcomes in terms of communication skills after EI.

[0046] The techniques described herein can be applied to predicting the effects of any intervention related to autism, including behavioral interventions, neural interventions, pharmacological interventions, etc. Some interventions can be indirect (e.g., training parents to provide therapy in the child's natural environment), and some interventions can be direct (e.g., administering a drug to the child). In some embodiments, the same input features or different subsets of input features can be used to train different machine learning models to predict the outcomes of different interventions (or different combinations of interventions). Machine learning models can also be trained using children who have not been given any intervention, and such models can predict future ASD-related diagnoses for children who have not had an intervention. For example, the prediction can involve the presence or severity of ASD, or the presence or severity of any feature related to autism, autistic-like behavior on the spectrum, or an autism subtype. Examples of such features include emotion and self-regulation; social interaction; social communication; language and communication; motor skills; executive function; cognition; and / or restricted or repetitive patterns of behavior, interests, or activities.

[0047] A variety of machine learning algorithms can be used (including any classifier or other algorithm that can be trained to predict the outcome of an unobserved data sample based on a set of data samples with known outcomes). In the example described above, a random forest model can be used to quantitatively predict improvements due to EI. Other models can be substituted, such as support vector regression (SVR), hidden Markov models, Bayesian classifiers, classifiers based on univariate or multivariate analysis, and deep learning algorithms (e.g., artificial neural networks). In various embodiments, the prediction can be quantitative or qualitative. For example, an SVM can provide a binary classifier that can be used to indicate whether a child is likely or unlikely to experience significant improvement due to EI. (For example, the minimum difference between the post-intervention and pre-intervention assessment scores can be defined as the threshold for significant improvement.) In cases where improvement is measured on an ordinal scale (e.g., high, medium, low, none), RankSVM or a similar model can allow prediction of the level of improvement. Other models can output numerical predictions, e.g., predicting the quantitative amount of improvement or the probability of significant improvement, etc. The parameters used to train and test the model can vary, including the size of the training dataset and the specific combination of input features. No specific algorithm implementation is required for training.

[0048] The training dataset can include children within a certain age range, e.g., from birth to 4 years old or from birth to 12 years old, etc.

[0049] The prediction results generated in the manner described herein can be used for treatment planning. For example, the prediction results can inform the decision of whether to continue an intervention for a given child. As another example, in cases where separate machine learning models can be trained to predict the effectiveness of different interventions (or combinations of interventions or no intervention), the predictions from an ensemble of models can inform the selection of an intervention for a given child. For example, the prediction based on no intervention can be compared with the predictions of various interventions.

[0050] All methods described herein are illustrative and can be modified. Within the scope permitted by logic, operations can be performed in an order different from the described order; the operations described above can be omitted or combined; and operations not explicitly described above can be added.

[0051] Although various circuits and components are described herein with reference to specific boxes, it should be understood that these boxes are defined for ease of description and are not intended to imply a particular physical arrangement of component parts. These boxes need not correspond to physically distinct components, and the same physical component may be used to implement aspects of multiple boxes. Components described as dedicated or fixed-function circuits may be configured to perform operations by providing an appropriate arrangement of circuit components (e.g., logic gates, registers, switches, etc.); automated design tools may be used to generate an appropriate arrangement of circuit components to implement the operations described herein. Components described as processors or microprocessors may be configured to perform the operations described herein by providing suitable program code. Depending on the manner in which the initial configuration is obtained, the individual boxes may be reconfigurable or non-reconfigurable. Embodiments of the present invention may be implemented in a variety of devices including electronic devices implemented using a combination of circuits and software.

[0052] Accordingly, although the present invention has been described with reference to specific embodiments, it should be understood that the invention is intended to cover all modifications and equivalents within the scope of the appended claims.

Claims

1. A method of predicting the outcome of an intervention for a child with actual or potential autism spectrum disorder (ASD), the method comprising: obtaining neural data characterizing one or both of anatomical properties or functional properties of at least a portion of a central nervous system of the child; extracting quantitative data from the neural data; Determine a baseline assessment of the child's ASD-related characteristics; determining a predicted change in the child's baseline assessment by analyzing the quantitative data using a machine learning model that has been trained to predict a change in an assessment of an ASD-related characteristic of the child, wherein the machine learning model is trained based on corresponding quantitative data and assessments of ASD-related characteristics from a training dataset obtained from a plurality of children who have previously had intervention outcomes; and Based on the baseline assessment and the predicted change in the baseline assessment, a predicted outcome of the intervention is generated.

2. The method of claim 1, wherein the neural data comprises electroencephalogram (EEG) data.

3. The method of claim 1, wherein the neural data comprises resting data and data obtained while the child is exposed to communicative stimulation.

4. The method of claim 1, wherein the neural data comprises one or more of the following: Electroencephalogram (EEG) data; Magnetic resonance imaging (MRI) data; functional magnetic resonance imaging (fMRI) data; Magnetoencephalography (MEG) data; or Functional near-infrared spectroscopy (fNIRS).

5. The method of claim 1, wherein the quantitative data comprises one or more of the following: First, functional data characterizes cortical or subcortical responses at rest; second functional data characterizing a cortical or subcortical response to the first stimulus; third functional data characterizing a cortical or subcortical response to a second stimulus that is different from the first stimulus; data indicating a difference between the first functional data and the second functional data; or Data indicating a difference between the second functional data and the third functional data.

6. The method of claim 1, wherein the quantitative data comprises one or more of the following: data characterizing one or more of the morphology, volume, or thickness of one or more regions of the child's central nervous system; data characterizing the integrity of white matter fiber tracts in one or more regions of the child's central nervous system; data characterizing a functional connection between two structures of the child's central nervous system; data characterizing structures along one or more cortical or subcortical pathways; Data characterizing function along neural pathways; Data characterizing the function of processing centers in sensory neural pathways; Data characterizing the function of brainstem nuclei and connecting pathways; or Data characterizing the function of thalamic nuclei and connecting pathways.

7. The method of claim 1, wherein the ASD-associated feature is a feature associated with autism, autistic-like behaviors on the spectrum, or a subtype of autism.

8. The method of claim 7, wherein the ASD-related features include one or more of the following: emotions and self-regulation; Social interaction; Social communication; Language and communication; Motor skills; Executive function; cognition; or Restrictive or repetitive patterns of behavior, interests, or activities.

9. The method of claim 1, wherein determining the baseline assessment comprises administering an assessment tool.

10. The method of claim 9, wherein the assessment tool is implemented online or offline in person by a natural person or a machine.

11. The method of claim 9, wherein the known outcomes in the training data set for a prior intervention for a child are determined by administering the assessment tool a second time after the intervention.

12. The method of claim 1, wherein the intervention comprises early intervention prior to a formal autism diagnosis.

13. The method of claim 1, wherein the intervention comprises one or more of the following: Behavioral therapy administered by professionals; behavioral therapy administered by parents to said children in a natural setting; Neurological intervention; or Pharmacological interventions.

14. The method of claim 1, wherein the intervention comprises allowing time to pass and the predicted outcome corresponds to a prediction of a future diagnosis of autism or ASD.

15. The method of claim 1, wherein the machine learning model comprises a random forest model.

16. The method of claim 1, wherein the machine learning model comprises a support vector machine (SVM).

17. The method of claim 1, wherein the machine learning model is based on Bayesian inference.

18. The method of claim 1, wherein the machine learning model is based on univariate or multivariate analysis.

19. The method of claim 1, wherein the prediction results include improved quantitative predictions.

20. The method of claim 1, wherein the prediction results include improved qualitative predictions.

21. The method of claim 20, wherein the qualitative prediction is a significant improvement or a not significant improvement.

22. The method of claim 1, further comprising: obtain birth, family and health data of said children; The child's birth, family and health data are input into the machine learning model together with the quantitative data.

23. The method of claim 1, wherein a score reflecting a baseline assessment of ASD-related characteristics of the child is input into the machine learning model along with the quantitative data.

24. The method of claim 1, further comprising: obtaining additional psychological data about the child, including scores on one or more additional assessments of ASD-related characteristics, Wherein, the additional psychological data is input into the machine learning model together with the quantitative data.

25. The method of claim 1, wherein the training data set comprises neural data obtained from the child from birth to 12 years of age and developmental assessment results obtained at an age of at least 4 months.

26. The method of claim 1, wherein the child is under 12 years of age and the training data is obtained from previous children under 12 years of age.

27. A system comprising: Memory; and A processor coupled to the memory and configured to perform the method as claimed in any one of claims 1 to 26.

Images