AI-Augmented Behavior Analysis for Children with Developmental Disabilities: Building Towards Precision Treatment
Shadi Ghafghazi, Amarie Carnett, Leslie Neely, Arun Das, Paul Rad
IIEEE 1
AI-Augmented Behavior Analysisfor Children with Developmental Disabilities:Building Towards Precision Treatment
Shadi Ghafghazi, Amarie Carnett, and Leslie Neely,
Child and Adolescent Policy and Research Institute (CAPRI)
Arun Das and Paul Rad,
Secure AI Laboratory for Autonomy (AILA)
Abstract —Autism spectrum disorder is a developmental dis-order characterized by significant social, communication, andbehavioral challenges. Individuals diagnosed with autism, intel-lectual, and developmental disabilities (AUIDD) typically requirelong-term care and targeted treatment and teaching. Effectivetreatment of AUIDD relies on efficient and careful behavioralobservations done by trained applied behavioral analysts (ABAs).However, this process overburdens ABAs by requiring the clini-cians to collect and analyze data, identify the problem behaviors,conduct pattern analysis to categorize and predict categoricaloutcomes, hypothesize responsiveness to treatments, and detectthe effects of treatment plans. Successful integration of digitaltechnologies into clinical decision-making pipelines and the ad-vancements in automated decision-making using Artificial Intel-ligence (AI) algorithms highlights the importance of augmentingteaching and treatments using novel algorithms and high-fidelitysensors. In this article, we present an AI-Augmented Learningand Applied Behavior Analytics (AI-ABA) platform to providepersonalized treatment and learning plans to AUIDD individuals.By defining systematic experiments along with automated datacollection and analysis, AI-ABA can promote self-regulativebehavior using reinforcement-based augmented or virtual realityand other mobile platforms. Thus, AI-ABA could assist cliniciansto focus on making precise data-driven decisions and increasethe quality of individualized interventions for individuals withAUIDD.
Autism spectrum disorder (“autism”) is a developmentaldisorder characterized by significant social, communication,and behavioral challenges. According to the National HealthStatistics Report [1], the prevalence of children aged 3-17years diagnosed with a developmental disability has increasedconsiderably from 6.99% in 2014-2016 to 17.8% in 2015-2018. Developmental disabilities can be severe, long-termdisorders often including intellectual impairments, physicalimpairments, or both. Intellectual disabilities, defined bysignificant limitations in cognition and adaptive functioning,are some of the most common impairments diagnosed duringthe developmental years, while physical impairments aretypically identifiable from birth. Often, these impairments co-occur and individuals diagnosed with autism, intellectual, anddevelopmental disabilities (AUIDD) typically require long-termcare and targeted treatment and teaching.The physical and mental limitations presenting due toAUIDD affect different aspects of life, including personal (self-care, independent living) and social skills (keeping conversa-tions, public speaking), often causing children to learn anddevelop slower than a typical child. While there is no cure for
Fig. 1. System architecture of the AI-augmented applied behavior analyticsplatforms is illustrated. Multimodal sensory information is collected usingboth invasive and non-invasive sensors which is processed by AI algorithmsto support decision making in treatment and learning paradigms of behavioranalysts. All data are stored securely in the cloud accessible by practitioners.Reinforcement paradigms are set up in a personalized fashion unique to eachindividuals. children with AUIDD, there are several types of treatments suchas applied behavioral analysis (ABA), occupational therapy,speech therapy, physical therapy, and pharmacological therapyavailable. Each of these treatments have proven effective inhelping individuals with AUIDD achieve a high level of skilldevelopment with earlier treatment leading to larger treatmentgains. Detecting and diagnosing these developmental disordersearly could help apply the needed treatment procedures andfacilitate many students’ learning and functioning to improvethe behavior of children over time.Effective treatment relies on efficient observation. However,the process of behavioral observation is time consumingrequiring clinicians to collect and analyze data [2], use thecollected data to identify the function of a problem behavior a r X i v : . [ c s . C Y ] F e b EEE 2 (or what the child is trying to communicate), use collected datato conduct pattern analysis to categorize problems and predictcategorical outcomes, use assessment data to hypothesizeresponsiveness to treatment, design treatment plans using ahypothesis regarding patient responsiveness to treatment, andcollect ongoing treatment data to detect the effects of thetreatment plan. For example, behavioral assessments used inABA, such as the functional behavioral assessment (FBA), areused to detect behavior patterns through indirect observations(screening instruments), direct observations and experimentalanalysis to identify behavioral function. The identified functionis then matched with a function-based behavioral intervention[3]. However, since clinicians rely on human collected data tomake decisions, chances of unreliable decisions are likely high,particularly considering varied clinical training that may resultin differences from one behavior analyst clinician to another.Recent research illustrates the importance of personalizedtreatment plans for individuals with AUIDD. ABA is aneffective treatment for a variety of developmental problems inchildren with developmental and intellectual disabilities andhas been successfully applied in different settings such ashome, school, and other educational centers. The beauty ofABA is that it allows the clinician to tailor interventions andpersonalize treatment to the situation and the individual’s need.Similarly, in education, having an individual education plan(IEP) is important for teachers to ensure that students receivespecialized instruction and related intervention. Personalizedtreatment and education plans are one of the most effectivetreatment for children with AUIDD. However, as previouslydiscussed, the process is time-consuming and unreliable dueto limitations with humans as the data collectors and patternanalyzers, reliance on professional judgement, and the everincreasing demands placed on behavior analysts and educators.While the treatment and education plans may be individualized,they are often not precise or efficient . In addition, time spentby clinicians collecting and analyzing the data often detractsfrom providing empathetic treatment. Given the high impact ofdigital learning platforms, artificial intelligence (AI), and cloudcomputing, designing and applying an appropriate assessmentframework for ABA and clinical interventions can assistclinicians and educators to effectively assess and monitor eachchild’s behavior and quickly modify interventions to meet hisor her specific needs and to account for various differencesacross environments.I. A
UGMENTING T EACHING AND P RECISION T REATMENT
Digital platforms and technologies are argued by many tohave a pivotal role in the dynamics of changing landscapeof clinical treatment. The main arguments include improvedsupport for treatment by 1) contextualizing and increasingmotivation of students and promoting engagement throughinteractive environments and reward structures, 2) providing alearning experience which caters to pace of patient’s individuallearning, and 3) providing continuous and life-long learningthrough mobile learning platforms [4].Several studies have shown promising results in improvingthe learning performance and boosting motivation to learn using graphical contents and interactions [5]. Teaching strategies andinterventions that utilize digital games in mobile devices ortablets have also shown promise for incorporating behaviorsmanagement techniques into games. Restrictive and repetitivebehaviors and interests (RRIBs) that might occur while playinggames could be monitored and treated using embedded auto-mated redirection to other games or levels to prevent interferingbehaviors that prevent access to learning opportunities andhelp promote calmness [6]. However, considering the range offunctioning for individuals with cognitive disabilities, furtherstudies involving precision of treatment options to ensureindividualization are needed, as well as replication of theseresults across large cohorts of participants [7].
A. Role of Augmented and Virtual Reality
Emerging technologies such as augmented reality (AR),including virtual reality (VR) and mixed reality (MR), is inthe forefront of recent technology-embedded practices thatoverlays reality and supplies additional layers to augment theperception of users [8] as well as enabling real time interactionof real and virtual objects [9]. Recent interest in using ARand VR technologies to aid adults and children with ASDprovides additional sensory information such as eye- trackingas well as a virtual platform to continuously interact withpeople and environment around them in a controlled settingwhile collecting data for future analysis.Safe and side-effect-free technologies are changing howAR/VR platforms are being found to be beneficial for improvingsoft skills, behaviors, and improving emotional skills [10].However, the importance of personalized services to provideaugmentation for individual learners has yet to be researchedat large. Thus, the need for personalized adaptive learningparadigms is required to improve engagement, autonomy, and topromote individualized preferences for children with cognitivedisorders. Data-driven algorithms could make use of thesedigital technologies to improve data collection while reducingthe demands placed on behavioral analysts and educators andthe time-required to collect and label these behavioral data.
B. Artificially Intelligent Methods in Behavioral Health
Recent advancements in artificial intelligence (AI) haveenabled real time human action performance [11], facialbehavioral analysis [12], speech analysis [13], speech disfluencydetection [14], stereotypical motor movement from sensorydata [15], many more. Published research from the last 5years shows the use of a wide variety of sensory inputs topredict human behavior, diseases, and cognitive states using AImethods, especially deep learning (DL). Electroencephalograms(EEG) has been used extensively to study the internal brainstates by recording the electrical activity of brain waves forpredicting diseases such as Parkinson’s [16]. Even thoughusing EEG to study Autism could have contradictions based onthe experimental conditions during EEG registration betweensubjects, age differences, and diversity of subjects, the abnormalEEG laterization in subjects with ASD can be leveraged tobuild AI models to predict traits of autism [17].
EEE 3
TABLE IS
UMMARY OF COMMON BEHAVIORAL CHALLENGES , THEIR DESCRIPTIONS , AND POSSIBLE MEASUREMENTS USING SENSORS
Behavioral Challenge Description Possible Sensors and MeasurementsCommunication Delays Children with neurodevelopmental disabilities, such as ASD, oftenrequire communication interventions. Additionally, approximately30% of individuals with ASD do not fully acquire spokencommunication [20]. Vocal communication could be more precisely evaluatedusing high tech data collection, which could promotemore accurate and reliable intervention procedures. Au-tomated data collection could be embedded in speech-generating device software along with isolated teachingplatforms to teach the use of these alternative communi-cation modes and promote data-based decision making.Problem Behavior End result of a chain of events that usually triggers (reactive, im-pulsive) during interactions with social and physical environmentsoften accompanied by anger or frustration. Problem behavior, suchas aggression, can be towards self, others, or the environment,while verbal aggression is usually towards others. Understandingthe underlying conditions and patterns are very important for BA’sto help reduce problem behavior. Problem behavior such as biting, hitting, kicking couldbe tracked using body tracking sensors and algorithms[11]. Vocal aggression such as screaming, shouting, useof foul language, etc. could be found by recording thespeech and studying them using intelligent algorithms.Agitation Restless or hyperactive stages, usually due to unrecognized physicalor sensory discomfort, which could lead to anxiety, verbal orphysical aggression. Physiological data from wearable devices and EEGsystems could help understand the long-term triggersand etiological factors leading to agitation.Vocal Stereotypy Persistent repetitions of words, phrases, or sounds without contex-tual or functional meaning in the current social setting. The vocal behavior could be found by recording thespeech and studying them using intelligent algorithmsin a functional analysis experimental setting [21].Sleep Disorder Trouble falling asleep or staying asleep for long periods of time.Could affect academic and social behaviors due to mood problems,memory, concentration, and learning problems, sluggish reactiontime, etc. caused by disrupted sleep patterns, insomnia, or sleepapnea [22]. Wearable sleep staging sensors or sleep sensing matscould identify sleep patterns before and after behavioraltherapies to study impact of the therapies in sleep andto design therapies that could improve sleep.Social Delay Difficulties with understanding social rules and interactions. Can berelated to communication delay and engagement in problem behav-ior but may also be present in individuals without communicationdelays. Specially designed games on AR/VR environments couldteach social skills and provide a structured environmentfor practicing vital social interactions.
Prior research using deep learning algorithms illustratethe successful use of facial videos collected using camerasto estimate the attention and engagement of children withdevelopmental disorders [18]. Similarly, inertial measurementunit (IMU) sensors which rely on accelerometer, gyroscope, andmagnetometers that can collect information about the frequency,intensity, and duration of physical activities have shown todetect stereotypical movements in ASD children [15]. As ourchildren rely more on digital devices such as iPads and mobilephones to read, learn, and interact from an early age [19], it isonly natural to research in facilitating access to individualizeddigital content through a variety of interfaces (such as games,interactive lessons, maps, and more) to cater for the exceptionallearners.
C. Challenges of AI in Behavioral Health
Despite the impressive role of AI in behavioral health, thereare two key open challenges which limits its use in clinical
Fig. 2. Tracking aggressive behavior using body tracking AI algorithm isillustrated. Here, forceful movements to the head can be classified as bangingthe head using a temporal action indexing module [11] as described in ourprevious research. decision-making. They are: 1) limited amount of labelled datato train AI algorithms and 2) black-box nature of deep neuralnetwork models. There are two potential solutions to theseproblems. Firstly, self-supervised representation learning hasbeen recently used to learn meaningful dense representationsfrom small amount of data. Also, reinforcement learningparadigms can learn to optimize based on an exploration-exploitation paradigm on any defined environment. Secondly,explainable artificial intelligence (XAI) methods can be usedto improve the transparency and trust of decision-making bygenerating meta-information to describe ‘why’ and ‘how’ adecision was made while suggesting ‘what’ features influencedthe decision the most [23].II. P
ERSONALIZED E XPLAINABLE AI TO I MPROVE
ABA
AND T REATMENT
Augmenting and complementing this systematic evaluation ofsubjects using intelligent algorithms is an avenue that warrantsfurther research. Recent literature highlights the importanceof online treatments and large adoption of telehealth solutions[24] across the globe. Building an Explainable AI-AugmentedLearning and Applied Behavior Analytics (AI-ABA) platformcould complement licensed behavior analysts and therapistswho rely on direct observation of audio-visual cues and otherphysiological data available during a session to diagnose andprovide feedback to subjects. AI-ABA could make use of facialexpressions, extremity movements, speech tone, heart-rate, andother available data to build automated pipelines to detect,diagnose, and alert BA’s of during treatment sessions withclients. Some of the desirable qualities of the AI-augmentedABA and learning platform are as follows:
EEE 4
Fig. 3. Attention span of children can be studied by capturing EEG signals or eye-tracking information whereas vocal stereotypy can be predicted by studyingthe facial muscle movement patterns over time. In (a), we illustrate a deep learning model generating a prediction and corresponding brain activation mapshighlighting the most attributing region of the brain. Vocal stereotypy (b) can be detected by observing the repetitive motion in the lower facial muscle groupsusing a facial muscle movement analyzer. • Ability to collect cognitive, perceptive, speech, movement,and physiological multimodal data of individuals. • Ability to define systematic experiments with dynamicreinforcers in virtual or physical environments. • Ability to manage experimental data and results ofrecurring experiments of individuals. • Provide early detection of behavioral changes in childrenand explain why, how, and what features were used forthe prediction.Moreover, AI-ABA platforms can promote self-regulativebehavior using reinforcement-based AR/VR or game environ-ments and collected multimodal data. It could also promotecreativity and curiosity by designing personalized dynamicenvironments which involve physical activities and speechinteractions while completing set tasks. By supporting gamesand virtual agents, AI-ABA could promote the development ofspoken communication and social communicative behavioursuch that the collaboration and social skills can be improved byinteracting with virtual agents in social settings using naturallanguage technologies.Figure 1 illustrates the system architecture of an AI-ABAparadigm which complements human intelligence with AIalgorithms. General system architecture of AI-ABA mustconsist of 1) multimodal sensing technologies to collect avariety of sensor data, 2) AI-augmented Treatment Personalizer(ATP) to ingest the sensor data and generate personalizedtreatment plans and IEPs to children, 3) suite of explainableAI algorithms to support various requirements of the ATP,and 4) integration to various presentation formats or front-endtechnologies such as AR/VR/MR [25] or tablets [26].
A. Behavioral Challenges and Sensing Technologies
Table I summarizes some of the common treated behavioralchallenges for children with AUIDD and possible sensor mea-surements that could improve data collection while reducingthe physical and mental burden of arduous observational datacollection for BA’s during the treatments. Agitation, aggression,and stereotypy seen in children with developmental disorderscould reoccur due to the process of reinforcement that parents unknowingly encourage in the household. Hence, it is veryimportant to track behavioral changes in multiple settings andenvironments to design individualized treatment plans or IEPsfor children.A problem behavior or challenging behavior is any culturallyabnormal behavior that could jeopardize the physical safetyof the individual or others that often restricts the personfrom social or communal activities. Self-harm or self-injuriousbehavior (SIB), harming caregivers, or general aggression isseen in children with low frustration tolerance. Throwing itemsat people, hitting themselves with objects, banging head againstan object, etc. are visible behaviors of aggression and agitationin children with problem behaviors. Audible and/or visualbehavior such as vocal stereotypy is commonly identified byparents as their children repeat words repetitively. However,despite the many intervention models described in research,only a few focuses on the impact of intervention model invocal stereotypy and the secondary impacts in other behaviors[27].Problem behaviors can be evaluated in multiple settings andenvironments by collecting multimodal sensor data includingEEG, gait and limb movements and body tracking usingcameras. Figure 2 illustrates one such method to index mildversus intense actions in humans. By generating a temporal mapof body movements, a spatio-temporal LSTM based attentionnetwork is used to highlight the areas of the body with rapidmovements to classify actions. A kinetic fuzzy intensity analysisnetwork generates an action intensity based on the temporalaction map and deep learning prediction. Attention and agitationcan be studied by a temporal analysis of the facial musclemovements and emotional states.Attention span can be studied using a temporal analysis ofEEG signals as illustrated in Figure 3 (a). Brain attention mapscan be generated using deep learning algorithms to understandthe pre and post brain states leading to a more detailedunderstanding than traditional methods. Vocal stereotypy can bedetected by observing the lower facial muscle movements forrepetitive motions. This is illustrated in Figure 3 (b). Here, themicroexpressions on the face leading to repeated behaviors andsounds can be used as inputs to a deep learning algorithm to
EEE 5
Fig. 4. Wearable physiological sensors such as Empatica E4 can be used tocollect heart rate data and study variability to carry out sleep staging and sleepquality analysis before and after therapy sessions. Here, the red dotted linesindicate intermittent waking up states. classify vocal stereotypy among other behaviors. We have foundMicrosoft Azure Kinect camera to be a well-rounded sensor torecord body tracking data, RGB, IR, and depth videos, and also,stereo audio data. Additionally, we found Empatica E4 wearablesensor to be reliable to study sleep disorders using heart rate,electrodermal activity, and other sensory data as illustrated inFigure 4. Integration of data collected via behavioral sensingcan lead to deeper understanding of the triggers of problembehavior, such as poor sleep the night before. It can also helpto distinguish behaviors triggered by environmental events,versus those rooted in biological processes, and facilitate earlyidentification of behaviors via precursors, such as heart rate.Precise understanding of behavioral triggers is essential toensuring effective and efficient treatment.Various other actions, such as pacing in a repeated circularmotion, body rocking, back and forth movement of fingers,tapping objects repeatedly, etc. are used as pointers to under-stand the hidden behaviors that might help with a diagnosis.Single-subject research designs (SSRDs), focused on individualparticipants, are often used, even now, to identify and evaluateinterventions due to the heterogeneity of the AUIDD population.However, integration of AI into ABA can allow for collectionof larger data sets and analysis of multi-modal behavioraldata, enhancing research standards and the strength of theresearch evidence. In reality, finding interventions that canbe considered evidence-based practices (EBPs) require largecohorts of subjects and multi-session datasets [28]. Hence,scalable architectures such as AI-ABA should be exploredas a means to collect data and to promote early detectionAUIDD from large subject populations and predictive modelsfacilitating responsiveness to intervention.III. T HE R OAD A HEAD
Systems such as AI-ABA could assists practitioners inmaking the precise data-based decisions to increase the qualityof individualized intervention for individuals with AUIDD.This type of system is likely to have profound effects on theimprovement in treatment efficacy and treatment outcomes, through a blend of both real-world and virtual elementsthat embeds generalization of targeted skills across multipleenvironments for the various aspects of an individual’s life.Recent studies illustrate the importance of active involvementof children in the intervention process [28], [29]. Additionally,providing an learning environment with effective interactionand communication is another important factor that should betaken into account [30], [31]. Since AI augmented ABA providethe optimal use of delivering ABA services for children withdevelopmental and cognitive needs, the system could easilybe integrated to across environments, such as schools andhome to deliver a more comprehensive treatment program.This smart and connected health via behavioral sensing wouldnot only be highly desirable for existing outpatient clinics, butcan be integrated into telehealth platforms to facilitate serviceaccess to those living in rural and deprived services areas, andcan provide a level of system resiliency during periods wherehuman interaction is limited, such as the ongoing COVID-19pandemic. In addition, applying home-based ABA can helpensure parents involvement with their child training alongsideother stakeholders (e.g., teachers, caregivers) to create a morecomprehensive support network for service delivery.AI-ABA will help researchers to focus on the precisionwithin intervention research, reinforcers effectiveness, andindividualized treatment models while augmenting part of thedata collection and analysis to AI algorithms. For example, asystem capable of handling multiple sensory inputs for datacapture could in a plug-and-play manner collect data usingspecific application programming interfaces (APIs) and processthem using existing machine learning algorithms. The processeddata and results could be used to dynamically influencethe virtual environments, learning structures, the precisionof treatment plans, effectiveness, and the incorporation ofAR or VR digital platforms to promote greater access tointervention and generalization of treatment effects. AR andVR feedback loops could increase learning engagement andraise comprehension of topics, provide interaction, improvecommunication, trigger imagination, and enhance problem-solving skills, especially when involving spatial skills [25]. Acombination of AR and VR technologies with other invasiveand non-invasive data collection systems could collect bothphysiological and behavioral data to study temporal dynamicsof behavior in children. Additionally, this could enable just-in-time adaptive interventions (JITAIs) by collecting precisedata and provide a repository of prior behavior of each client.A
CKNOWLEDGMENT
This project was funded partly by the Open Cloud Institute(OCI) at University of Texas at San Antonio (UTSA) andpartly by the UTSA Brain Health Consortium and Officeof the Vice President for Research, Economic Development,and Knowledge Enterprise. Arun Das and Shadi Ghafgazhicontributed equally. The authors gratefully acknowledge theuse of the services of Jetstream cloud.A
BOUT THE A UTHORS
Shadi Ghafghazi ([email protected]) is currentlya Senior Lecturer at the University of Applied Science and
EEE 6
Technology where she is teaching special needs educationcourses. She earned her Master’s degree in Information Scienceand Knowledge Studies from Shahid Beheshti University andreaped her Bachelor’s degree in Library and InformationScience from University of Tehran, Iran.
Amarie Carnett ([email protected]) is a SeniorLecturer with the College of Education, Victoria Universityof Wellington, New Zealand and a Research Faculty at theUniversity of Texas at San Antonio. She is a doctorate levelBoard Certified Behavior Analyst (BCBA-D).
Leslie Neely ([email protected]) is an Associate Pro-fessor in the College of Education and Human Development,University of Texas at San Antonio and Director of the Childand Adolescent Policy and Research Institute. She is a doctoratelevel Board Certified Behavior Analyst (BCBA-D).
Arun Das ([email protected]) is currently a Ph.D. can-didate at UT San Antonio, TX, USA where he focuses onexplainable AI and self-supervised algorithms for healthcareand neuroscience domains. He earned his MS in ComputerEngineering from UT San Antonio, TX, USA and his B.Tech.from Cochin University of Science and Technology, Cochin,India. Arun is a graduate student member of the IEEE LoneStar section and the IEEE Eta Kappa Nu honor society.
Paul Rad ([email protected]) is an Associate Professorwith the Department of Computer Science, The University ofTexas at San Antonio. He is a Senior Member of the NationalAcademy of Inventors and a Senior member of IEEE.R
EFERENCES[1] B. Zablotsky and L. I. Black, “Prevalence of children aged 3–17 yearswith developmental disabilities, by urbanicity: United states, 2015–2018,”
National Health Statistics Reports; no 139. , 2020.[2] J. J. Love, C. F. Miguel, J. K. Fernand, and J. K. LaBrie, “The effects ofmatched stimulation and response interruption and redirection on vocalstereotypy,”
Journal of Applied Behavior Analysis , vol. 45, no. 3, pp.549–564, 2012.[3] F. M. Gresham, “Establishing the technical adequacy of functional behav-ioral assessment: Conceptual and measurement challenges,”
BehavioralDisorders , vol. 28, no. 3, pp. 282–298, 2003.[4] W. Ng, “Change and continuity in educational uses of new digitaltechnologies,” in
New digital technology in education . Springer, 2015,pp. 3–23.[5] J. Bacca, S. Baldiris, R. Fabregat, S. Graf et al. , “Augmented realitytrends in education: a systematic review of research and applications,”
Journal of Educational Technology & Society , vol. 17, no. 4, pp. 133–149,2014.[6] S. Alarcon-Licona, L. Loke, and N. Ahmadpour, “From autism educatorsto game designers: Integrating teaching strategies into game designfor autism education support,” in
Proceedings of the 30th AustralianConference on Computer-Human Interaction , 2018, pp. 58–62.[7] G. B. Schaefer, “Clinical genetic aspects of autism spectrum disorders,”
International journal of molecular sciences , vol. 17, no. 2, p. 180, 2016.[8] R. Azuma, Y. Baillot, R. Behringer, S. Feiner, S. Julier, and B. MacIntyre,“Recent advances in augmented reality,”
IEEE computer graphics andapplications , vol. 21, no. 6, pp. 34–47, 2001.[9] T. D. Parsons, A. R. Carlew, J. Magtoto, and K. Stonecipher, “Thepotential of function-led virtual environments for ecologically valid mea-sures of executive function in experimental and clinical neuropsychology,”
Neuropsychological rehabilitation , vol. 27, no. 5, pp. 777–807, 2017.[10] N. T. Sahin, N. U. Keshav, J. P. Salisbury, and A. Vahabzadeh, “Safetyand lack of negative effects of wearable augmented-reality socialcommunication aid for children and adults with autism,”
Journal ofclinical medicine , vol. 7, no. 8, p. 188, 2018.[11] N. Bendre, N. Ebadi, J. J. Prevost, and P. Najafirad, “Human actionperformance using deep neuro-fuzzy recurrent attention model,”
IEEEAccess , vol. 8, pp. 57 749–57 761, 2020. [12] T. Baltrusaitis, A. Zadeh, Y. C. Lim, and L.-P. Morency, “Openface2.0: Facial behavior analysis toolkit,” in . IEEE,2018, pp. 59–66.[13] N. Memari, S. Abdollahi, S. Khodabakhsh, S. Rezaei, and M. Moghbel,“Speech analysis with deep learning to determine speech therapy forlearning difficulties,” in
International Conference on Intelligent andFuzzy Systems . Springer, 2020, pp. 1164–1171.[14] A. Das, J. Mock, H. Chacon, F. Irani, E. Golob, and P. Najafirad,“Stuttering speech disfluency prediction using explainable attributionvectors of facial muscle movements,” arXiv preprint arXiv:2010.01231 ,2020.[15] N. M. Rad, S. M. Kia, C. Zarbo, T. van Laarhoven, G. Jurman,P. Venuti, E. Marchiori, and C. Furlanello, “Deep learning for automaticstereotypical motor movement detection using wearable sensors in autismspectrum disorders,”
Signal Processing , vol. 144, pp. 180–191, 2018.[16] S. L. Oh, Y. Hagiwara, U. Raghavendra, R. Yuvaraj, N. Arunkumar,M. Murugappan, and U. R. Acharya, “A deep learning approach forparkinson’s disease diagnosis from eeg signals,”
Neural Computing andApplications , pp. 1–7, 2018.[17] B. Penchina, A. Sundaresan, S. Cheong, and A. Martel, “Deep lstmrecurrent neural network for anxiety classification from eeg in adolescentswith autism,” in
International Conference on Brain Informatics . Springer,2020, pp. 227–238.[18] A. Di Nuovo, D. Conti, G. Trubia, S. Buono, and S. Di Nuovo, “Deeplearning systems for estimating visual attention in robot-assisted therapyof children with autism and intellectual disability,”
Robotics , vol. 7, no. 2,p. 25, 2018.[19] R. Linke, T. Kothe, and F. Alt, “Tabooga: A hybrid learning app to supportchildren’s reading motivation,” in
Proceedings of the 2017 Conferenceon Interaction Design and Children , 2017, pp. 278–285.[20] E. L. Wodka, P. Mathy, and L. Kalb, “Predictors of phrase and fluentspeech in children with autism and severe language delay,”
Pediatrics ,vol. 131, no. 4, pp. e1128–e1134, 2013.[21] O. Healy, S. Lydon, T. Brady, M. Rispoli, J. Holloway, L. Neely, andI. Grey, “The use of differential reinforcement of other behavioursto establish inhibitory stimulus control for the management of vocalstereotypy in children with autism,”
Developmental neurorehabilitation ,vol. 22, no. 3, pp. 192–202, 2019.[22] A. Carnett, S. Hansen, L. McLay, L. Neely, and R. Lang, “Quantitative-analysis of behavioral interventions to treat sleep problems in childrenwith autism,”
Developmental neurorehabilitation , vol. 23, no. 5, pp.271–284, 2020.[23] A. Das and P. Rad, “Opportunities and challenges in explainable artificialintelligence (xai): A survey,” arXiv preprint arXiv:2006.11371 , 2020.[24] L. Neely, E. R. Hong, S. Kawamini, I. Umana, and I. Kurz, “Interconti-nental telehealth to train japanese interventionists in incidental teachingfor children with autism,”
Journal of Behavioral Education , pp. 1–16,2020.[25] A. Bilyalova, D. Salimova, and T. Zelenina, “Digital transformation ineducation,” in
International Conference on Integrated Science . Springer,2019, pp. 265–276.[26] N. Kucirkova, “ipads in early years, separating assumptions and evidence,”
Frontiers in Psychology , vol. 5, 2014.[27] W. H. Ahearn, K. M. Clark, R. P. MacDonald, and B. I. Chung, “Assessingand treating vocal stereotypy in children with autism,”
Journal of appliedbehavior analysis , vol. 40, no. 2, pp. 263–275, 2007.[28] S. J. Singh, T. Iacono, and K. M. Gray, “Interactions of pre-symbolicchildren with developmental disabilities with their mothers and siblings,”
International Journal of Language & Communication Disorders , vol. 50,no. 2, pp. 202–214, 2015.[29] A. Alias and N. M. Salleh, “Analysis of problems faced by specialeducation teacher in teaching the multiple disabilities students,”
Journalof ICSAR , vol. 1, no. 1, pp. 60–67, 2017.[30] P. S¸afak, M. E. C¸ etin, and M. Kot, “Siblings attitudes towards personswith severe/profound and multiple disabilities in turkey,”
Procedia-Socialand Behavioral Sciences , vol. 191, pp. 2083–2088, 2015.[31] M. Syrj¨am¨aki, N. Sajaniemi, E. Suhonen, A. Alijoki, and M. Nislin,“Enhancing peer interaction: an aspect of a high-quality learningenvironment in finnish early childhood special education,”