Use of Machine Learning to Identify Children with Autism and Their Motor Abnormalities

Applying Machine Learning to Kinematic and Eye Movement Features of a Movement Imitation Task to Predict Autism Diagnosis

Autism is a developmental condition currently identified by experts using observation, interview, and questionnaire techniques and primarily assessing social and communication deficits. Motor function and movement imitation are also altered in autism and can be measured more objectively. In this study, motion and eye tracking data from a movement imitation task were combined with supervised machine learning methods to classify 22 autistic and 22 non-autistic adults. The focus was on a reliable machine learning application. We have used nested validation to develop models and further tested the models with an independent data sample. Feature selection was aimed at selection stability to assure result interpretability. Our models predicted diagnosis with 73% accuracy from kinematic features, 70% accuracy from eye movement features and 78% accuracy from combined features. We further explored features which were most important for predictions to better understand movement imitation differences in autism. Consistent with the behavioural results, most discriminative features were from the experimental condition in which non-autistic individuals tended to successfully imitate unusual movement kinematics while autistic individuals tended to fail. Machine learning results show promise that future work could aid in the diagnosis process by providing quantitative tests to supplement current qualitative ones.

Machine Learning and Virtual Reality on Body Movements' Behaviors to Classify Children with Autism Spectrum Disorder

Autism spectrum disorder (ASD) is mostly diagnosed according to behavioral symptoms in sensory, social, and motor domains. Improper motor functioning, during diagnosis, involves the qualitative evaluation of stereotyped and repetitive behaviors, while quantitative methods that classify body movements' frequencies of children with ASD are less addressed. Recent advances in neuroscience, technology, and data analysis techniques are improving the quantitative and ecological validity methods to measure specific functioning in ASD children. On one side, cutting-edge technologies, such as cameras, sensors, and virtual reality can accurately detect and classify behavioral biomarkers, as body movements in real-life simulations. On the other, machine-learning techniques are showing the potential for identifying and classifying patients' subgroups. Starting from these premises, three real-simulated imitation tasks have been implemented in a virtual reality system whose aim is to investigate if machine-learning methods on movement features and frequency could be useful in discriminating ASD children from children with typical neurodevelopment. In this experiment, 24 children with ASD and 25 children with typical neurodevelopment participated in a multimodal virtual reality experience, and changes in their body movements were tracked by a depth sensor camera during the presentation of visual, auditive, and olfactive stimuli. The main results showed that ASD children presented larger body movements than TD children, and that head, trunk, and feet represent the maximum classification with an accuracy of 82.98%. Regarding stimuli, visual condition showed the highest accuracy (89.36%), followed by the visual-auditive stimuli (74.47%), and visual-auditive-olfactory stimuli (70.21%). Finally, the head showed the most consistent performance along with the stimuli, from 80.85% in visual to 89.36% in visual-auditive-olfactory condition. The findings showed the feasibility of applying machine learning and virtual reality to identify body movements' biomarkers that could contribute to improving ASD diagnosis.

Towards the automatic detection of social biomarkers in autism spectrum disorder: introducing the simulated interaction task (SIT)

Social interaction deficits are evident in many psychiatric conditions and specifically in autism spectrum disorder (ASD), but hard to assess objectively. We present a digital tool to automatically quantify biomarkers of social interaction deficits: the simulated interaction task (SIT), which entails a standardized 7-min simulated dialog via video and the automated analysis of facial expressions, gaze behavior, and voice characteristics. In a study with 37 adults with ASD without intellectual disability and 43 healthy controls, we show the potential of the tool as a diagnostic instrument and for better description of ASD-associated social phenotypes. Using machine-learning tools, we detected individuals with ASD with an accuracy of 73%, sensitivity of 67%, and specificity of 79%, based on their facial expressions and vocal characteristics alone. Especially reduced social smiling and facial mimicry as well as a higher voice fundamental frequency and harmony-to-noise-ratio were characteristic for individuals with ASD. The time-effective and cost-effective computer-based analysis outperformed a majority vote and performed equal to clinical expert ratings.

Machine Learning to Study Social Interaction Difficulties in ASD

Autism Spectrum Disorder (ASD) is a spectrum of neurodevelopmental conditions characterized by difficulties in social communication and social interaction as well as repetitive behaviors and restricted interests. Prevalence rates have been rising, and existing diagnostic methods are both extremely time and labor consuming. There is an urgent need for more economic and objective automatized diagnostic tools that are independent of language and experience of the diagnostician and that can help deal with the complexity of the autistic phenotype. Technological advancements in machine learning are offering a potential solution, and several studies have employed computational approaches to classify ASD based on phenomenological, behavioral or neuroimaging data. Despite of being at the core of ASD diagnosis and having the potential to be used as a behavioral marker for machine learning algorithms, only recently have movement parameters been used as features in machine learning classification approaches. In a proof-of-principle analysis of data from a social interaction study we trained a classification algorithm on intrapersonal synchrony as an automatically and objectively measured phenotypic feature from 29 autistic and 29 typically developed individuals to differentiate those individuals with ASD from those without ASD. Parameters included nonverbal motion energy values from 116 videos of social interactions. As opposed to previous studies to date, our classification approach has been applied to non-verbal behavior objectively captured during naturalistic and complex interactions with a real human interaction partner assuring high external validity. A machine learning approach lends itself particularly for capturing heterogeneous and complex behavior in real social interactions and will be essential in developing automatized and objective classification methods in ASD.

Artificial intelligence for precision medicine in neurodevelopmental disorders

The ambition of precision medicine is to design and optimize the pathway for diagnosis, therapeutic intervention, and prognosis by using large multidimensional biological datasets that capture individual variability in genes, function and environment. This offers clinicians the opportunity to more carefully tailor early interventions- whether treatment or preventative in nature-to each individual patient. Taking advantage of high performance computer capabilities, artificial intelligence (AI) algorithms can now achieve reasonable success in predicting risk in certain cancers and cardiovascular disease from available multidimensional clinical and biological data. In contrast, less progress has been made with the neurodevelopmental disorders, which include intellectual disability (ID), autism spectrum disorder (ASD), epilepsy and broader neurodevelopmental disorders. Much hope is pinned on the opportunity to quantify risk from patterns of genomic variation, including the functional characterization of genes and variants, but this ambition is confounded by phenotypic and etiologic heterogeneity, along with the rare and variable penetrant nature of the underlying risk variants identified so far. Structural and functional brain imaging and neuropsychological and neurophysiological markers may provide further dimensionality, but often require more development to achieve sensitivity for diagnosis. Herein, therefore, lies a precision medicine conundrum: can artificial intelligence offer a breakthrough in predicting risks and prognosis for neurodevelopmental disorders? In this review we will examine these complexities, and consider some of the strategies whereby artificial intelligence may overcome them.

The Use of Artificial Intelligence in Screening and Diagnosis of Autism Spectrum Disorder: A Literature Review

Objectives:

The detection of autism spectrum disorder (ASD) is based on behavioral observations. To build a more objective datadriven method for screening and diagnosing ASD, many studies have attempted to incorporate artificial intelligence (AI) technologies. Therefore, the purpose of this literature review is to summarize the studies that used AI in the assessment process and examine whether other behavioral data could potentially be used to distinguish ASD characteristics.

Methods:

Based on our search and exclusion criteria, we reviewed 13 studies.

Results:

To improve the accuracy of outcomes, AI algorithms have been used to identify items in assessment instruments that are most predictive of ASD. Creating a smaller subset and therefore reducing the lengthy evaluation process, studies have tested the efficiency of identifying individuals with ASD from those without. Other studies have examined the feasibility of using other behavioral observational features as potential supportive data.

Conclusion:

While previous studies have shown high accuracy, sensitivity, and specificity in classifying ASD and non-ASD individuals, there remain many challenges regarding feasibility in the real-world that need to be resolved before AI methods can be fully integrated into the healthcare system as clinical decision support systems.

A Novel Virtual Sample Generation Method to Overcome the Small Sample Size Problem in Computer Aided Medical Diagnosing

Deep neural networks are successful learning tools for building nonlinear models. However, a robust deep learning-based classification model needs a large dataset. Indeed, these models are often unstable when they use small datasets. To solve this issue, which is particularly critical in light of the possible clinical applications of these predictive models, researchers have developed approaches such as virtual sample generation. Virtual sample generation significantly improves learning and classification performance when working with small samples. The main objective of this study is to evaluate the ability of the proposed virtual sample generation to overcome the small sample size problem, which is a feature of the automated detection of a neurodevelopmental disorder, namely autism spectrum disorder. Results show that our method enhances diagnostic accuracy from 84%–95% using virtual samples generated on the basis of five actual clinical samples. The present findings show the feasibility of using the proposed technique to improve classification performance even in cases of clinical samples of limited size. Accounting for concerns in relation to small sample sizes, our technique represents a meaningful step forward in terms of pattern recognition methodology, particularly when it is applied to diagnostic classifications of neurodevelopmental disorders. Besides, the proposed technique has been tested with other available benchmark datasets. The experimental outcomes showed that the accuracy of the classification that used virtual samples was superior to the one that used original training data without virtual samples.

Kinematic features of a simple and short movement task to predict autism diagnosis

Autism is a developmental condition primarily identified by social and communication deficits. However, over 70% of autistic individuals also show motor function deficits, which are evident even when simple stereotyped movements are performed. In this study, we have asked 24 autistic and 22 non-autistic adults to perform pointing movements between two markers 30 cm apart as quickly and as accurately as they can for 10 seconds. Motion tracking was employed to collect data and calculate kinematic features of the movement and aiming accuracy. At the group level, the results showed that autistic individuals performed pointing movements slower but more accurately compared to non-autistic individuals. At the individual level, we have used Machine Learning methods to predict autism diagnosis. Nested result Cross-Validation was used, which in contrast to commonly used K-fold Cross-Validation avoids pooling training and testing data and provides robust performance estimates. Our developed models achieved a statistically significant classification accuracy of 71% and showed that even a simple and short motor task enables discrimination between autistic and non-autistic individuals.

The Heterogeneity Problem: Approaches to Identify Psychiatric Subtypes

The imprecise nature of psychiatric nosology restricts progress towards characterizing and treating mental health disorders. One issue is the 'heterogeneity problem': different causal mechanisms may relate to the same disorder, and multiple outcomes of interest can occur within one individual. Our review tackles this heterogeneity problem, providing considerations, concepts, and approaches for investigators examining human cognition and mental health. We highlight the difficulty of pure dimensional approaches due to 'the curse of dimensionality'. Computationally, we consider supervised and unsupervised statistical approaches to identify putative subtypes within a population. However, we emphasize that subtype identification should be linked to a particular outcome or question. We conclude with novel hybrid approaches that can identify subtypes tied to outcomes, and may help advance precision diagnostic and treatment tools.

Infant Motor Development in Autism Spectrum Disorder: A Synthesis and Meta-analysis

Studies of infant motor development in autism spectrum disorder (ASD) have increased in recent years. This article synthesized this literature through meta-analysis to assess (a) whether infant motor ability differs in ASD relative to neurotypical controls; and (b) whether motor ability and communication are related in infants with ASD. Study 1 aggregated data from 1,953 infants with ASD (ages 3.0-42.0 months), and Study 2 included 890 infants with ASD (age 6.0-42.9 months). Study 1 revealed that infant motor ability differed significantly in ASD compared with neurotypical infants-this difference was robust to variation in measurement and design. Furthermore, this group difference amplified as age increased. Study 2 indicated that within ASD, infant motor ability and communication are related.

Objective measurement of head movement differences in children with and without autism spectrum disorder

Background

Deficits in motor movement in children with autism spectrum disorder (ASD) have typically been characterized qualitatively by human observers. Although clinicians have noted the importance of atypical head positioning (e.g. social peering and repetitive head banging) when diagnosing children with ASD, a quantitative understanding of head movement in ASD is lacking. Here, we conduct a quantitative comparison of head movement dynamics in children with and without ASD using automated, person-independent computer-vision based head tracking (Zface). Because children with ASD often exhibit preferential attention to nonsocial versus social stimuli, we investigated whether children with and without ASD differed in their head movement dynamics depending on stimulus sociality.

Methods

The current study examined differences in head movement dynamics in children with (n = 21) and without ASD (n = 21). Children were video-recorded while watching a 16-min video of social and nonsocial stimuli. Three dimensions of rigid head movement—pitch (head nods), yaw (head turns), and roll (lateral head inclinations)—were tracked using Zface. The root mean square of pitch, yaw, and roll was calculated to index the magnitude of head angular displacement (quantity of head movement) and angular velocity (speed).

Results

Compared with children without ASD, children with ASD exhibited greater yaw displacement, indicating greater head turning, and greater velocity of yaw and roll, indicating faster head turning and inclination. Follow-up analyses indicated that differences in head movement dynamics were specific to the social rather than the nonsocial stimulus condition.

Conclusions

Head movement dynamics (displacement and velocity) were greater in children with ASD than in children without ASD, providing a quantitative foundation for previous clinical reports. Head movement differences were evident in lateral (yaw and roll) but not vertical (pitch) movement and were specific to a social rather than nonsocial condition. When presented with social stimuli, children with ASD had higher levels of head movement and moved their heads more quickly than children without ASD. Children with ASD may use head movement to modulate their perception of social scenes.

Electronic supplementary material

The online version of this article (10.1186/s13229-018-0198-4) contains supplementary material, which is available to authorized users.

Subtyping cognitive profiles in Autism Spectrum Disorder using a Functional Random Forest algorithm

DSM-5 Autism Spectrum Disorder (ASD) comprises a set of neurodevelopmental disorders characterized by deficits in social communication and interaction and repetitive behaviors or restricted interests, and may both affect and be affected by multiple cognitive mechanisms. This study attempts to identify and characterize cognitive subtypes within the ASD population using our Functional Random Forest (FRF) machine learning classification model. This model trained a traditional random forest model on measures from seven tasks that reflect multiple levels of information processing. 47 ASD diagnosed and 58 typically developing (TD) children between the ages of 9 and 13 participated in this study. Our RF model was 72.7% accurate, with 80.7% specificity and 63.1% sensitivity. Using the random forest model, the FRF then measures the proximity of each subject to every other subject, generating a distance matrix between participants. This matrix is then used in a community detection algorithm to identify subgroups within the ASD and TD groups, and revealed 3 ASD and 4 TD putative subgroups with unique behavioral profiles. We then examined differences in functional brain systems between diagnostic groups and putative subgroups using resting-state functional connectivity magnetic resonance imaging (rsfcMRI). Chi-square tests revealed a significantly greater number of between group differences (p < .05) within the cingulo-opercular, visual, and default systems as well as differences in inter-system connections in the somato-motor, dorsal attention, and subcortical systems. Many of these differences were primarily driven by specific subgroups suggesting that our method could potentially parse the variation in brain mechanisms affected by ASD.

The Utility of a Computerized Algorithm Based on a Multi-Domain Profile of Measures for the Diagnosis of Attention Deficit/Hyperactivity Disorder

The current gold standard for diagnosis of attention deficit/hyperactivity disorder (ADHD) includes subjective measures, such as clinical interview, observation, and rating scales. The significant heterogeneity of ADHD symptoms represents a challenge for this assessment and could prevent an accurate diagnosis. The aim of this work was to investigate the ability of a multi-domain profile of measures, including blood fatty acid (FA) profiles, neuropsychological measures, and functional measures from near-infrared spectroscopy (fNIRS), to correctly recognize school-aged children with ADHD. To answer this question, we elaborated a supervised machine-learning method to accurately discriminate 22 children with ADHD from 22 children with typical development by means of the proposed profile of measures. To assess the performance of our classifier, we adopted a nested 10-fold cross validation, where the original dataset was split into 10 subsets of equal size, which were used repeatedly for training and testing. Each subset was used once for performance validation. Our method reached a maximum diagnostic accuracy of 81% through the combining of the predictive models trained on neuropsychological, FA profiles, and deoxygenated-hemoglobin features. With respect to the analysis of a single-domain dataset per time, the most discriminant neuropsychological features were measures of vigilance, focused and sustained attention, and cognitive flexibility; the most discriminating blood FAs were linoleic acid and the total amount of polyunsaturated fatty acids. Finally, with respect to the fNIRS data, we found a significant advantage of the deoxygenated-hemoglobin over the oxygenated-hemoglobin data in terms of predictive accuracy. These preliminary findings show the feasibility and applicability of our machine-learning method in correctly identifying children with ADHD based on multi-domain data. The present machine-learning classification approach might be helpful for supporting the clinical practice of diagnosing ADHD, even fostering a computer-aided diagnosis perspective.

Applying machine learning to identify autistic adults using imitation: An exploratory study

Autism spectrum condition (ASC) is primarily diagnosed by behavioural symptoms including social, sensory and motor aspects. Although stereotyped, repetitive motor movements are considered during diagnosis, quantitative measures that identify kinematic characteristics in the movement patterns of autistic individuals are poorly studied, preventing advances in understanding the aetiology of motor impairment, or whether a wider range of motor characteristics could be used for diagnosis. The aim of this study was to investigate whether data-driven machine learning based methods could be used to address some fundamental problems with regard to identifying discriminative test conditions and kinematic parameters to classify between ASC and neurotypical controls. Data was based on a previous task where 16 ASC participants and 14 age, IQ matched controls observed then imitated a series of hand movements. 40 kinematic parameters extracted from eight imitation conditions were analysed using machine learning based methods. Two optimal imitation conditions and nine most significant kinematic parameters were identified and compared with some standard attribute evaluators. To our knowledge, this is the first attempt to apply machine learning to kinematic movement parameters measured during imitation of hand movements to investigate the identification of ASC. Although based on a small sample, the work demonstrates the feasibility of applying machine learning methods to analyse high-dimensional data and suggest the potential of machine learning for identifying kinematic biomarkers that could contribute to the diagnostic classification of autism.

From movement kinematics to social cognition: the case of autism

The way in which we move influences our ability to perceive, interpret and predict the actions of others. Thus movements play an important role in social cognition. This review article will appraise the literature concerning movement kinematics and motor control in individuals with autism, and will argue that movement differences between typical and autistic individuals may contribute to bilateral difficulties in reciprocal social cognition.

Toward the Autism Motor Signature: Gesture patterns during smart tablet gameplay identify children with autism

Autism is a developmental disorder evident from infancy. Yet, its clinical identification requires expert diagnostic training. New evidence indicates disruption to motor timing and integration may underpin the disorder, providing a potential new computational marker for its early identification. In this study, we employed smart tablet computers with touch-sensitive screens and embedded inertial movement sensors to record the movement kinematics and gesture forces made by 37 children 3-6 years old with autism and 45 age- and gender-matched children developing typically. Machine learning analysis of the children's motor patterns identified autism with up to 93% accuracy. Analysis revealed these patterns consisted of greater forces at contact and with a different distribution of forces within a gesture, and gesture kinematics were faster and larger, with more distal use of space. These data support the notion disruption to movement is core feature of autism, and demonstrate autism can be computationally assessed by fun, smart device gameplay.

Use of machine learning to improve autism screening and diagnostic instruments: effectiveness, efficiency, and multi-instrument fusion

BACKGROUND

Machine learning (ML) provides novel opportunities for human behavior research and clinical translation, yet its application can have noted pitfalls (Bone et al., 2015). In this work, we fastidiously utilize ML to derive autism spectrum disorder (ASD) instrument algorithms in an attempt to improve upon widely used ASD screening and diagnostic tools.

METHODS

The data consisted of Autism Diagnostic Interview-Revised (ADI-R) and Social Responsiveness Scale (SRS) scores for 1,264 verbal individuals with ASD and 462 verbal individuals with non-ASD developmental or psychiatric disorders, split at age 10. Algorithms were created via a robust ML classifier, support vector machine, while targeting best-estimate clinical diagnosis of ASD versus non-ASD. Parameter settings were tuned in multiple levels of cross-validation.

RESULTS

The created algorithms were more effective (higher performing) than the current algorithms, were tunable (sensitivity and specificity can be differentially weighted), and were more efficient (achieving near-peak performance with five or fewer codes). Results from ML-based fusion of ADI-R and SRS are reported. We present a screener algorithm for below (above) age 10 that reached 89.2% (86.7%) sensitivity and 59.0% (53.4%) specificity with only five behavioral codes.

CONCLUSIONS

ML is useful for creating robust, customizable instrument algorithms. In a unique dataset comprised of controls with other difficulties, our findings highlight the limitations of current caregiver-report instruments and indicate possible avenues for improving ASD screening and diagnostic tools.

Identifying children with autism spectrum disorder based on their face processing abnormality: A machine learning framework

The atypical face scanning patterns in individuals with Autism Spectrum Disorder (ASD) has been repeatedly discovered by previous research. The present study examined whether their face scanning patterns could be potentially useful to identify children with ASD by adopting the machine learning algorithm for the classification purpose. Particularly, we applied the machine learning method to analyze an eye movement dataset from a face recognition task [Yi et al., 2016], to classify children with and without ASD. We evaluated the performance of our model in terms of its accuracy, sensitivity, and specificity of classifying ASD. Results indicated promising evidence for applying the machine learning algorithm based on the face scanning patterns to identify children with ASD, with a maximum classification accuracy of 88.51%. Nevertheless, our study is still preliminary with some constraints that may apply in the clinical practice. Future research should shed light on further valuation of our method and contribute to the development of a multitask and multimodel approach to aid the process of early detection and diagnosis of ASD. Autism Res 2016, 9: 888-898. © 2016 International Society for Autism Research, Wiley Periodicals, Inc.

Multivariate analyses applied to fetal, neonatal and pediatric MRI of neurodevelopmental disorders

Multivariate analysis (MVA) is a class of statistical and pattern recognition methods that involve the processing of data that contains multiple measurements per sample. MVA can be used to address a wide variety of medical neuroimaging-related challenges including identifying variables associated with a measure of clinical importance (i.e. patient outcome), creating diagnostic tests, assisting in characterizing developmental disorders, understanding disease etiology, development and progression, assisting in treatment monitoring and much more. Compared to adults, imaging of developing immature brains has attracted less attention from MVA researchers. However, remarkable MVA research growth has occurred in recent years. This paper presents the results of a systematic review of the literature focusing on MVA technologies applied to neurodevelopmental disorders in fetal, neonatal and pediatric magnetic resonance imaging (MRI) of the brain. The goal of this manuscript is to provide a concise review of the state of the scientific literature on studies employing brain MRI and MVA in a pre-adult population. Neurological developmental disorders addressed in the MVA research contained in this review include autism spectrum disorder, attention deficit hyperactivity disorder, epilepsy, schizophrenia and more. While the results of this review demonstrate considerable interest from the scientific community in applications of MVA technologies in pediatric/neonatal/fetal brain MRI, the field is still young and considerable research growth remains ahead of us.