Classification Of Autism Research Articles

A residual graph convolutional network with spatio-temporal features for autism classification from fMRI brain images

Diagnosing autism spectrum disorder (ASD) remains a challenge due to its complexity and insufficient evidence available for its diagnosis. Recent research in the field of psychiatry suggests that there are no clear causes for ASD, but a hypothesis is raised that abnormalities in the superior temporal sulcus (STS), which is connected to visual cortex regions, can serve as a critical indicator of ASD. Inspired by this hypothesis, this paper proposes a deep learning model with two parts to diagnose ASD by utilizing functional brain connectivity between STS and visual cortex. The first part is a residual attention network that selectively extracts the structural and temporal features from 4D brain images while maintaining dynamic connectivity between the two regions. The second part is a graph convolutional network that determines ASD from the graph with 39 nodes constructed by the residual attention network. Experiments with the fMRI data from 800 patients known as ABIDE (Autism Brain Imaging Data Exchange) and 10-fold cross-validation show that the proposed model outperforms the state-of-the-art methods by achieving 11.37%p improvement in the ASD classification. Additional analyses justify each part of the proposed model through an ablation study and various visualizations.

Early Identification of Autism Spectrum Disorder Among Children Aged 4 Years — Autism and Developmental Disabilities Monitoring Network, 11 Sites, United States, 2020

Autism spectrum disorder (ASD). 2020. The Autism and Developmental Disabilities Monitoring Network is an active surveillance program that estimates prevalence and characteristics of ASD and monitors timing of ASD identification among children aged 4 and 8 years. In 2020, a total of 11 sites (located in Arizona, Arkansas, California, Georgia, Maryland, Minnesota, Missouri, New Jersey, Tennessee, Utah, and Wisconsin) conducted surveillance of ASD among children aged 4 and 8 years and suspected ASD among children aged 4 years. Surveillance included children who lived in the surveillance area at any time during 2020. Children were classified as having ASD if they ever received 1) an ASD diagnostic statement in an evaluation, 2) a special education classification of autism (eligibility), or 3) an ASD International Classification of Diseases (ICD) code (revisions 9 or 10). Children aged 4 years were classified as having suspected ASD if they did not meet the case definition for ASD but had a documented qualified professional's statement indicating a suspicion of ASD. This report focuses on children aged 4 years in 2020 compared with children aged 8 years in 2020. For 2020, ASD prevalence among children aged 4 years varied across sites, from 12.7 per 1,000 children in Utah to 46.4 in California. The overall prevalence was 21.5 and was higher among boys than girls at every site. Compared with non-Hispanic White children, ASD prevalence was 1.8 times as high among Hispanic, 1.6 times as high among non-Hispanic Black, 1.4 times as high among Asian or Pacific Islander, and 1.2 times as high among multiracial children. Among the 58.3% of children aged 4 years with ASD and information on intellectual ability, 48.5% had an IQ score of ≤70 on their most recent IQ test or an examiner's statement of intellectual disability. Among children with a documented developmental evaluation, 78.0% were evaluated by age 36 months. Children aged 4 years had a higher cumulative incidence of ASD diagnosis or eligibility by age 48 months compared with children aged 8 years at all sites; risk ratios ranged from 1.3 in New Jersey and Utah to 2.0 in Tennessee. In the 6 months before the March 2020 COVID-19 pandemic declaration by the World Health Organization, there were 1,593 more evaluations and 1.89 more ASD identifications per 1,000 children aged 4 years than children aged 8 years received 4 years earlier. After the COVID-19 pandemic declaration, this pattern reversed: in the 6 months after pandemic onset, there were 217 fewer evaluations and 0.26 fewer identifications per 1,000 children aged 4 years than children aged 8 years received 4 years earlier. Patterns of evaluation and identification varied among sites, but there was not recovery to pre-COVID-19 pandemic levels by the end of 2020 at most sites or overall. For 2020, prevalence of suspected ASD ranged from 0.5 (California) to 10.4 (Arkansas) per 1,000 children aged 4 years, with an increase from 2018 at five sites (Arizona, Arkansas, Maryland, New Jersey, and Utah). Demographic and cognitive characteristics of children aged 4 years with suspected ASD were similar to children aged 4 years with ASD. A wide range of prevalence of ASD by age 4 years was observed, suggesting differences in early ASD identification practices among communities. At all sites, cumulative incidence of ASD by age 48 months among children aged 4 years was higher compared with children aged 8 years in 2020, indicating improvements in early identification of ASD. Higher numbers of evaluations and rates of identification were evident among children aged 4 years until the COVID-19 pandemic onset in 2020. Sustained lower levels of ASD evaluations and identification seen at a majority of sites after the pandemic onset could indicate disruptions in typical practices in evaluations and identification for health service providers and schools through the end of 2020. Sites with more recovery could indicate successful strategies to mitigate service interruption, such as pivoting to telehealth approaches for evaluation. From 2016 through February of 2020, ASD evaluation and identification among the cohort of children aged 4 years was outpacing ASD evaluation and identification 4 years earlier (from 2012 until March 2016) among the cohort of children aged 8 years in 2020 . From 2016 to March 2020, ASD evaluation and identification among the cohort of children aged 4 years was outpacing that among children aged 8 years in 2020 from 2012 until March 2016. The disruptions in evaluation that coincided with the start of the COVID-19 pandemic and the increase in prevalence of suspected ASD in 2020 could have led to delays in ASD identification and interventions. Communities could evaluate the impact of these disruptions as children in affected cohorts age and consider strategies to mitigate service disruptions caused by future public health emergencies.

An Evaluation of a Teaching Interaction Procedure Implemented in a Recess Setting

The teaching interaction procedure (TIP) is a strategy that has been demonstrated as effective in promoting social skill acquisition in school settings for young students with social communication deficits (Leaf et al., 2009; Leaf et al., 2010). However, a frequently cited criticism of social skills training is the lack of generalizability of target skills to novel contexts (Bellini et al., 2007). The purpose of the study was to evaluate a TIP-based social skills intervention conducted on the playground, intended to promote generalizability through training in naturalistic settings and to evaluate generalizability of skill acquisition to the classroom. Eight students 5-8 years old with an educational classification of autism or developmental delay participated in the study. The primary dependent variable was skill acquisition in the playground setting, and a secondary measure was generalized skill acquisition to the classroom setting. Target skills included appropriate body language, participation, and responding to initiations. A multi-probe design embedded within a multiple baseline design across target skills with concurrent replication across participants was used to evaluate the primary and secondary measures. Overall, results suggest that increases in skill acquisition were observed during implementation of the TIP across most participants and skills in both training and generalization phases. However, substantial variability was noted across participants related to maintaining skill acquisition during maintenance and follow-up phases in both the training and generalization settings. Limitations of these results are discussed as well as implications for school practitioners.

Application of TBSS-based machine learning models in the diagnosis of pediatric autism.

To explore the microstructural changes of white matter in children with pediatric autism by using diffusion kurtosis imaging (DKI), and evaluate whether the combination of tract-based spatial statistics (TBSS) and back-propagation neural network (BPNN)/support vector machine (SVM)/logistic regression (LR) was feasible for the classification of pediatric autism. DKI data were retrospectively collected from 32 children with autism and 27 healthy controls (HCs). Kurtosis fractional anisotropy (FAK), mean kurtosis (MK), axial kurtosis (KA), radial kurtosis (RK), fractional anisotropy (FA), axial diffusivity (DA), mean diffusivity (MD) and Radial diffusivity (DR) were generated by iQuant workstation. TBSS was used to detect the regions of parameters values abnormalities and for the comparison between these two groups. In addition, we also introduced the lateralization indices (LI) to study brain lateralization in children with pediatric autism, using TBSS for additional analysis. The parameters values of the differentiated regions from TBSS were then calculated for each participant and used as the features in SVM/BPNN/LR. All models were trained and tested with leave-one-out cross validation (LOOCV). Compared to the HCs group, the FAK, DA, and KA values of multi-fibers [such as the bilateral superior longitudinal fasciculus (SLF), corticospinal tract (CST) and anterior thalamic radiation (ATR)] were lower in pediatric autism group (p < 0.05, TFCE corrected). And we also found DA lateralization abnormality in Superior longitudinal fasciculus (SLF) (the LI in HCs group was higher than that in pediatric autism group). However, there were no significant differences in FA, MD, MK, DR, and KR values between HCs and pediatric autism group (P > 0.05, TFCE corrected). After performing LOOCV to train and test three model (SVM/BPNN/LR), we found the accuracy of BPNN (accuracy = 86.44%) was higher than that of LR (accuracy = 76.27%), but no different from SVM (RBF, accuracy = 81.36%; linear, accuracy = 84.75%). Our proposed method combining TBSS findings with machine learning (LR/SVM/BPNN), was applicable in the classification of pediatric autism with high accuracy. Furthermore, the FAK, DA, and KA values and Lateralization index (LI) value could be used as neuroimaging biomarkers to discriminate the children with pediatric autism or not.

Improving Multi-Site Autism Classification via Site-Dependence Minimization and Second-Order Functional Connectivity.

Machine learning has been widely used to develop classification models for autism spectrum disorder (ASD) using neuroimaging data. Recently, studies have shifted towards using large multi-site neuroimaging datasets to boost the clinical applicability and statistical power of results. However, the classification performance is hindered by the heterogeneous nature of agglomerative datasets. In this paper, we propose new methods for multi-site autism classification using the Autism Brain Imaging Data Exchange (ABIDE) dataset. We firstly propose a new second-order measure of functional connectivity (FC) named as Tangent Pearson embedding to extract better features for classification. Then we assess the statistical dependence between acquisition sites and FC features, and take a domain adaptation approach to minimize the site dependence of FC features to improve classification. Our analysis shows that 1) statistical dependence between site and FC features is statistically significant at the 5% level, and 2) extracting second-order features from neuroimaging data and minimizing their site dependence can improve over state-of-the-art (SOTA) classification results, achieving a classification accuracy of 73%. The code is available at https://github.com/kundaMwiza/fMRI-site-adaptation.

Deep Learning for the Classification of Autism Using Functional Neuroimaging

Deep learning models have advanced many branches of science. However, these models have not been adequately developed for neuroimaging applications mainly because of the non-availability of large labelled datasets. In this study, we present an explainable deep learning approach to investigate the neurobiology of the autism spectrum disorder (ASD), which is one of the most prevalent neurodevelopmental disorders. Our approach achieved state of the art classification accuracy and identified brain features in discriminating ASDs from the typical subjects and finally identified features that predicted the severity of the symptoms.

Do labels matter? Analysis of teachers' self‐efficacy towards the autism spectrum disorder depending on the diagnostic label used (ASD or Asperger's)

Changes in the classification of autism and Asperger's syndrome led to changes in social perception of ASD. Since last criteria, studies indicate higher levels of stigma towards ASD than towards Asperger's. These prejudices are barriers to inclusive education. Thus, it is relevant (1) to evaluate pre‐service teachers' self‐efficacy towards the label of ASD; (2) to evaluate pre‐service teachers' self‐efficacy towards the label of Asperger's and (3) to compare those results to analyse whether the use of different diagnostic labels brings about different levels of self‐efficacy. One hundred and eighty‐six primary education pre‐service teachers participated in the current study. Two adaptations of the Autism Self‐Efficacy Scale for Teachers (ASSET) were used: a version with the label of ‘ASD’ (n = 96) and another for ‘Asperger's’ (n = 90). The scores obtained by the group asked about ASD were high according to the ASSET score range, while the scores obtained by the group asked about Asperger's were medium. After comparing the results, participants asked about the label ASD showed higher levels of self‐efficacy than participants asked about Asperger's. These results could be a consequence of the consolidation of the ASD diagnosis among society and the higher presence of children with ASD in schools and cultural products, among other factors.

Classification of Autism and Control Gait in Children Using Multisegment Foot Kinematic Features.

Previous research has demonstrated that children with autism walk with atypical ankle kinematics and kinetics. Although these studies have utilized single-segment foot (SSF) data, multisegment foot (MSF) kinematics can provide further information on foot mechanics. Machine learning (ML) tools allow the combination of MSF kinematic features for classifying autism gait patterns. In this study, multiple ML models are investigated, and the most contributing features are determined. This study involved 19 children with autism and 21 age-matched controls performing walking trials. A 34-marker system and a 12-camera motion capture system were used to compute SSF and MSF angles during walking. Features extracted from these foot angles and their combinations were used to develop support vector machine (SVM) models. Additional techniques-S Hapley Additive exPlanations (SHAP) and the Shapley Additive Global importancE (SAGE) are used for local and global importance of the black-box ML models. The results suggest that models based on combinations of MSF kinematic features classify autism patterns with an accuracy of 96.3%, which is higher than using SSF kinematic features (83.8%). The relative angle between the metatarsal and midfoot segments had the highest contribution to the classification of autism gait patterns. The study demonstrated that kinematic features from MSF angles, supported by ML models, can provide an accurate and interpretable classification of autism and control patterns in children.

Replication study for ADOS-2 cut-offs to assist evaluation of autism spectrum disorder.

The Autism Diagnostic Observation Schedule, Second Edition (ADOS-2) has been widely used for ASD assessment. While prior studies investigated sensitivity and specificity of ADOS-2 Modules 1-3, there has been limited research addressing algorithm cut-off scores to optimize ADOS-2 classification. The goal of this study was to assess algorithm cut-off scores for diagnosing ASD with Modules 1-3, and to evaluate alignment of the ADOS-2 classification with the best estimate clinical diagnosis. Participants included 3144 children aged 31 months or older who received ADOS-2 Modules 1-3, as well as the best estimate clinical diagnosis. Five classification statistics were reported for each module: sensitivity, specificity, positive predictive value, negative predictive value, and accuracy (i.e., Receiver Operator Classification Statistic), and these statistics were calculated for the optimal cut-off score. Frequency tables were used to compare ADOS-2 classification and the best estimate clinical diagnosis. Half of the sample received Module 3, 21% received Module 2, and 29% received Module 1. The overall prevalence of ASD was 60%; the male-to-female ratio was 4:1, and half of the sample was non-White. Across all modules, the autism spectrum cut-off score from the ADOS-2 manual resulted in high sensitivity (95%+) and low specificity (63%-73%). The autism cut-off score resulted in better specificity (76%-86%) with favorable sensitivity (81%-94%). The optimal cut-off scores for all modules based on the current sample were within the autism spectrum classification range except Module 2 Algorithm 2. In the No ASD group, 29% had false positives (ADOS-2 autism spectrum classification or autism classification). The ADOS-2 autism spectrum classification did not indicate directionality for diagnostic outcome (ASD 56% vs. No ASD 44%). While cut-off scores of ADOS-2 Modules 1-3 in the manual yielded good clinical utility in ASD assessment, false positives and low predictability of the autism spectrum classification remain challenging for clinicians.

ConVnet BiLSTM for ASD Classification on EEG Brain Signal

As a neurodevelopmental disability, Autism Spectrum Disorder (ASD) is classified as a spectrum disorder. The availability of an automated technology system to classify the ASD trait would have a significant impact on paediatricians, as it would assist them in diagnosing ASD in children using a quantifiable method. In this paper, we propose a novel autism diagnosis method that is based on a hybrid of the deep learning algorithms. This hybrid consists of a convolutional neural network (ConVnet) architecture that merges two LSTM blocks (BiLSTM) with the other direction of propagation to classify the output state on the brain signal data from electroencephalogram (EEG) on individuals; typically development (TD) and autism (ASD) obtained from the Simon Foundation Autism Research Initiative (SFARI) database to classify the output state. For a 70:30 data distribution, an accuracy of 97.7 percent was achieved. Proposed methods outperformed the current state-of-the art in terms of autism classification efficiency and have the potential to make a significant contribution to neuroscience research, as demonstrated by the results.

Adaptation, Content Validity and Reliability of the Autism Classification System of Functioning for Social Communication: From Toddlerhood to Adolescent-Aged Children with Autism.

The Autism Classification System of Functioning: Social Communication (ACSF) describes social communication functioning levels. First developed for preschoolers with ASD, this study tests an expanded age range (2-to-18years). The ACFS rates the child's typical and best (i.e., capacity) performance. Qualitative methods tested parent and clinician perspectives of the ACSF age expansion using content analysis. The ACSF was used twice by parents and professionals for the same child/youth. Reliabilities were assessed using weighted kappa. Content validity supported the ACSF's applicability, clarity, and usability. The ACSF adaptations did not change its original construct. Reliability were calculated from 90 parent and professional Time-1 and Time-2 ratings for children/youth (2.1-15.6years). Results showed good-to-very good intra-rater agreement (typical) and good inter-rater agreement (capacity).

Construct Validity of the Autism Classification System of Functioning: Social Communication (ACSF:SC) Across Childhood and Adolescence.

This study examined the construct validity of the Autism Classification System of Functioning: Social Communication (ACSF). Participants included 145 parents of children with autism (2-19 years). The degree of convergent and discriminant validity between parent reported ACSF and subscales from Social Responsiveness Scale 2nd edition and Behavior Assessment System for Children, 3rd Edition were examined against a priori hypotheses. We examined construct validity in the entire sample as well as in specific age cohorts. Our findings suggest that ACSF can provide a valid classification system of social communication ability in children with autism 2-19 years of age, and its two subscales may be used to examine different aspects of social communication ability.

A Virtual Reality Based System for the Screening and Classification of Autism.

Autism - also known as Autism Spectrum Disorders or Autism Spectrum Conditions - is a neurodevelopmental condition characterized by repetitive behaviours and differences in communication and social interaction. As a consequence, many autistic individuals may struggle in everyday life, which sometimes manifests in depression, unemployment, or addiction. One crucial problem in patient support and treatment is the long waiting time to diagnosis, which was approximated to thirteen months on average. Yet, the earlier an intervention can take place the better the patient can be supported, which was identified as a crucial factor. We propose a system to support the screening of Autism Spectrum Disorders based on a virtual reality social interaction, namely a shopping experience, with an embodied agent. During this everyday interaction, behavioral responses are tracked and recorded. We analyze this behavior with machine learning approaches to classify participants from an autistic participant sample in comparison to a typically developed individuals control sample with high accuracy, demonstrating the feasibility of the approach. We believe that such tools can strongly impact the way mental disorders are assessed and may help to further find objective criteria and categorization.

An architectural framework for automatic detection of autism using deep convolution networks and genetic algorithm

The brainchild in any medical image processing lied in how accurately the diseases are diagnosed. Especially in the case of neural disorders such as autism spectrum disorder (ASD), accurate detection was still a challenge. Several noninvasive neuroimaging techniques provided experts information about the functionality and anatomical structure of the brain. As autism is a neural disorder, magnetic resonance imaging (MRI) of the brain gave a complex structure and functionality. Many machine learning techniques were proposed to improve the classification and detection accuracy of autism in MRI images. Our work focused mainly on developing the architecture of convolution neural networks (CNN) combining the genetic algorithm. Such artificial intelligence (AI) techniques were very much needed for training as they gave better accuracy compared to traditional statistical methods.

Deep Learning for Autism Diagnosis and Facial Analysis in Children.

In this paper, we introduce a deep learning model to classify children as either healthy or potentially having autism with 94.6% accuracy using Deep Learning. Patients with autism struggle with social skills, repetitive behaviors, and communication, both verbal and non-verbal. Although the disease is considered to be genetic, the highest rates of accurate diagnosis occur when the child is tested on behavioral characteristics and facial features. Patients have a common pattern of distinct facial deformities, allowing researchers to analyze only an image of the child to determine if the child has the disease. While there are other techniques and models used for facial analysis and autism classification on their own, our proposal bridges these two ideas allowing classification in a cheaper, more efficient method. Our deep learning model uses MobileNet and two dense layers to perform feature extraction and image classification. The model is trained and tested using 3,014 images, evenly split between children with autism and children without it; 90% of the data is used for training and 10% is used for testing. Based on our accuracy, we propose that the diagnosis of autism can be done effectively using only a picture. Additionally, there may be other diseases that are similarly diagnosable.

Investigation of Hybrid Feature Selection Techniques for Autism Classification using EEG Signals

Investigation of Hybrid Feature Selection Techniques for Autism Classification using EEG Signals

Influences of student race/ethnicity and gender on autism special education classification considerations.

Autism prevalence has continued to rise in recent years. However, females and children from Black, Latinx, and Asian backgrounds are often misidentified or identified less often than White males. These identification disparities make it difficult for children to receive appropriate special education and school intervention services. In this study, school psychologists read a vignette featuring a student with possible autism symptoms. The vignette varied by student race/ethnicity and gender. Afterward, participants rated the likelihood that they would classify the student with autism and their confidence in this rating. Student race/ethnicity and gender influenced both classification likelihood and confidence. These results suggest that school psychologists are influenced by implicit bias and do not fully consider cultural factors in school autism evaluations. This may contribute to identification disparities.

LSTM-Based Electroencephalogram Classification on Autism Spectrum Disorder

Autism Spectrum Disorder (ASD) is categorized as a neurodevelopmental disability. Having an automated technology system to classify the ASD trait would have a huge influence on paediatricians, which can aid them in diagnosing ASD in children using a quantifiable method. A novel autism diagnosis method based on a bidirectional long-short-term-memory (LSTM) network's deep learning algorithm is proposed. This multi-layered architecture merges two LSTM blocks with the other direction of propagationto classify the output state on the brain signal datafrom an electroencephalogram (EEG) on individuals; normal and autism obtained from the Simon Foundation Autism Research Initiative (SFARI) database. The accuracy of 99.6% obtained for 90:10 train:test data distribution, while the accuracy of 97.3% was achieved for 70:30 distribution. The result shows that the proposed approach had better autism classification with upgraded efficiency compared to single LSTM network methodand potentially giving a significant contribution in neuroscience research.

LSTM-based Electroencephalogram Classification on Autism Spectrum Disorder

Autism Spectrum Disorder (ASD) is categorized as a neurodevelopmental disability. Having an automated technology system to classify the ASD trait would have a huge influence on paediatricians, which can aid them in diagnosing ASD in children using a quantifiable method. A novel autism diagnosis method based on a bidirectional long-short-term-memory (LSTM) network's deep learning algorithm is proposed. This multi-layered architecture merges two LSTM blocks with the other direction of propagation to classify the output state on the brain signal datafrom an electroencephalogram (EEG) on individuals; normal and autism obtained from the Simon Foundation Autism Research Initiative (SFARI) database. The accuracy of 99.6% obtained for 90:10 train: test data distribution, while the accuracy of 97.3% was achieved for 70:30 distribution. The result shows that the proposed approach had better autism classification with upgraded efficiency compared to single LSTM network method and potentially giving a significant contribution in neuroscience research

Preliminary psychometric properties of the Autism Classification System of Functioning: Social Communication (ACSF: SC) in Farsi speaking children

Background Autism Classification System of Functioning: Social Communication (ACSF: SC) assesses typical performance and capacity of communication. Farsi version of ACSF: SC was developed and its psychometric properties investigated. Material and methods The English version of ACSF: SC was translated into Farsi and cultural adaptation was completed. The content validity of the adapted version was evaluated by experts and parents. The back-translated version was approved by www.canchild.ca. Therapists and parents assessed face validity. Correlation between ACSF: SC and the Children’s Communication Checklist (CCC) was analyzed to determine concurrent validity. For inter-rater reliability, 5 speech and language pathologists, 5 occupational therapists, and 35 parents participated to assess the communication level of 35 autistic children aged 3–6. The test–retest reliability was performed at the interval of 14 days. Result The results of analyses for typical performance and capacity showed the communication levels of children in ACSF: CS were directly related to subscales of CCC including inappropriate initiation (r ≥ 0.46) and stereotypical conversation (r ≥ 0.45). The weighted kappa inter-rater reliability ranged from good (0.54) to excellent (0.77); and the test–retest reliability varied from good (0.63) to excellent (0.85). Conclusion Adapted version of ACSF: CS is able to estimate the communication function of Farsi- speaking autism spectrum disorder children.