2.2 Relevant literature on the problem

[sapphirachan]

Ethics

Screening methods Q-Chat-10

Other non-clinical screening methods

• AQ10
• EQ
• Eyes Test
• Faux Pas Test
• M-CHAT

Method of delivery Electronic versus non-electronic Other Mobile Apps

Machine Learning

[sapphirachan]

Sapphira

• Toddlers with ASD more successful with visual search
• Language Assessment and Development in Toddlers with Autism Spectrum Disorders.
• Screening for Autism Spectrum Disorders with the Brief Infant-Toddler Social and Emotional Assessment
• Comparative Analysis of Three Screening Instruments for Autism Spectrum Disorder in Toddlers at High Risk
• Mobile app for autism screening yields useful data
• Cognoa Begins Clinical Study to Support Increased Parent Engagement in Early Detection and Early Intervention of Developmental Delays
• Is this the clearest autism test ever? (PDQ-1)
• Rutgers professor heads study into onset detection of Autism
• Brief Questionnaire Aids Autism Diagnosis
• Development of a Machine Learning Algorithm for the Surveillance of Autism Spectrum Disorder
• Machine Learning; Investigators at University of Haifa Target Machine Learning (Risk Assessment for Parents Who Suspect Their Child Has Autism Spectrum Disorder: Machine Learning Approach)
• Age at First Identification of Autism Spectrum Disorder: An Analysis of Two US Surveys

[sapphirachan]

Franco

• Smartphone Apps for Autism Spectrum Disorder
• Developing Autism Screening Criteria for the Brief Infant Toddler Social Emotional Assessment (BITSEA)
• A Prospective Study of Response to Name in Infants at Risk for Autism
• A Pictorial Mobile Application for Improving Communication Skills in Non-Verbal Autism.
• Using Machine Learning for Detection of Autism Spectrum Disorder

[sapphirachan]

Machine Learning Algorithm for ASD / Classification to predict ASD Given the non-feasibility to determine ASD on a large scale, classification techniques are used on words and phrases that were extracted verbatim from evaluation, predicting with 86.5% accuracy to clinicians; determined. This was done by identifying the consistencies in description of symptoms recorded during evaluation. This method used random forest classifier. Factors for consideration when training the model is whether or not the performance will be impacted when used on a geographically or demographically different dataset without compromise on the performance and quality. However this method allows for the identification of ASD from unstructured text data. (Maenner, Yeargin-Allsop, Braun, Christensen & Schieve, 2016)

Maenner, Yeargin-Allsop, Braun, Christensen & Schieve. (2016). Development of a Machine Learning Algorithm for the Surveillance of Autism Spectrum Disorder. PLoS ONE, 11(12): e0168224. doi:10.1371/journal.pone.0168224

[sapphirachan]

Impact Of Age, Gender and Socio-Economic Background on Diagnosis Success and Challenges Despite recommendation by The American Academy of Paediatricians for ASD screenings to be done at 18 and 24 months of age, only 50% of children are screened at this age. According to a study by NCHS data brief, only 20% of diagnosed children was done by the time they were two years old, while more than 50% were diagnosed only when they reached the school going age of 5 or older.

Early detection of autism is challenging due to the inability to pinpoint a specific reliable method based on behaviour and observation. However this is crucial to ensure that the child receives the necessary help. Prior to the development of PDQ-10, it was also more challenging to diagnose females, as the tests were built around males. PDQ-10 test was developed based on a simple 10 questions that can be completed in 2 minutes, with an accuracy of 88% in identification, as found by researchers at Rutgers University, which was consistent across children from all types of socio-economic backgrounds. (Baulkman, 2018)

Baulkman, J. (2018, 7 Feb) Is this the clearest autism test ever? Two-minute questionnaire could detect the disorder in children earlier than anything else Retrieved fromhttp://www.dailymail.co.uk/health/article-5358897/A-two-minute-survey-detect-child-autism.html

[sapphirachan]

App to detect autism in children successful with video analysis algorithm Observation of children in their natural environment using video to detect their emotional response enables scalability for testing and analysing by highly trained professionals. This is conducted by playing various clips on the mobile phone screen and capturing their emotions via facial responses, which is possible since autistic children expressed their joy less frequently. There is potential for the child’s development to be monitored. The usage rate of this tool has proven its feasibility, with 88 % of the 4,441 uploaded videos being usable, from the participating 1,756 families participating over the course of one year. (Mobile App for Autism Screening Yields Useful Data, 2018)

(2018, June). Mobile App For Autism Screening Yields Useful Data. Mental Health Weekly Digest. Retrieved from: http://www.newsrx.com/newsletters/MentalHealthWeeklyDigest. html

[flfguerrero]

Machine Learning for ASD detectiton and prediction of severity.

Various screening methods have been created in order to identify the potential symptoms and behavioural traits related to ASD (Austism Spectrum Disorder). However without proper training and experience with qualified medical professionals to appropiately diagnose the presence of ASD, these identifiers and results cannot provide a definitve diagnosis to establish the presence of ASD within toddlers. Bekerom's (2017), research into whether machine learning could be used to identify ASD without the need for medical professional concluded the possibility of correctly predict the presence of ASD within toddlers exists. Using the J48 screening method with identifiable traits (developmental delay, learning disability, speech etc.) as his attributes, Bekoram's research found his machine learning algorithim resulted in 54.1% accuracy.

Bekoram's research extends beyond identifying ASD within toddlers and uses machine learning to predict the severity of ASD. Utilising the 1-away method, Bekoram classifies his cases as having No ASD, Mild ASD, Moderate ASD and Severe ASD. When applied to his J48 screening found a substantial increase in accuracy from 54.1% to 90.2%

Using Machine Learning for Detection of Autism Spectrum Disorder Bram van den Bekerom