TEENAGERS | ACCESSIBILITY | COMPUTER VISION | FACIAL LANDMARKS | DIGITAL BIOMARKERS
COMPANION
DESIGNING A PRE-SCREENING SERVICE
FOR UNDETECTED TEENAGE AUTISTIC INDIVIDUALS
ROLE
Research Team Lead & Qualitative Interviewer
DESCRIPTION
I led the user research and usability testing process while developing the prototype of a smartphone application to help potential teenage autistic individuals to check their likelihood of autism before getting an official diagnosis.
TEAMMATES
Sihyun Lee, Suvi Valtonen-Kim, Solveig Helena Asbjornsen
TIMELINE
Mar 2025 - Jun 2025
RESEARCH METHODS
Semi-Structured interviews, Think Aloud Protocol
TARGET USERS
Teenagers (13-19) who want to check their likelihood of being autistic
TOOLS
Figma
PROJECT INTRODUCTION
Companion is a prototype based on the idea of using Computer Vision and Machine Learning to detect signs of autism.
This application is aimed to provide a preliminary screening service so that teenage users can check if they have Autism Spectrum Disorder (ASD) or not, before getting an official diagnosis.
After going through games and a questionnaire, users can view their likelihood of having autism, as well as resources that may help them.
FINAL PROTOTYPE DEMO
7th iteration
PROBLEM STATEMENT
Social Stigma and Diagnostic Barriers Can Delay Identification of Autism
According to the WHO, around 1 in 127 persons around the world have autism. Cases of autistic individuals receiving a diagnosis later in life is quite frequent—some are diagnosed in their forties and fifties.
Undetected autism in teens can lead to mental health issues and social difficulties.
Centers for Disease Control and Prevention. (2025, May 26). Data and statistics on autism spectrum disorder. https://www.cdc.gov/autism/data-research/index.html
Zablotsky, B., Black, L. I., Maenner, M. J., Schieve, L. A., Danielson, M. L., Bitsko, R. H., Blumberg, S. J., Kogan, M. D., & Boyle, C. A. (2019). Prevalence and trends of developmental disabilities among children in the United States: 2009-2017. Pediatrics, 144(4), e20190811. https://doi.org/10.1542/peds.2019-0811
World Health Organization. (2025, September 16). Autism spectrum disorders. https://www.who.int/news-room/fact-sheets/detail/autism-spectrum-disorders
Stagg, S. D., & Belcher, H. (2019). Living with autism without knowing: receiving a diagnosis in later life. Health Psychology and Behavioral Medicine, 7(1), 348–361. https://doi.org/10.1080/21642850.2019.1684920
Hanna, J., Gates, J. A., & Hull, L. (2022). Mental health and social difficulties of late-diagnosed autistic adults: A narrative review. Autism, 26(3), 555–570. https://doi.org/10.1177/1362361320979182
What Is Autism?
Autism, or Autism Spectrum Disorder (ASD), is defined as a range of conditions characterized by challenges with social skills, repetitive behaviors, speech, and nonverbal communication.
Studies point to how many patients are left undiagnosed until much later, until adolescence or adulthood. This indicates how those patients are not able to get timely help.
It can be difficult to get a diagnosis early on for those who do not show signs of clear autism.
Medicine Unlocked, 31, 100974. https://doi.org/10.1016/j.imu.2022.100974
American Psychiatric Association. (n.d.). What is autism spectrum disorder?. https://www.psychiatry.org/patients-families/autism/what-is-autism-spectrum-disorder
Centers for Disease Control and Prevention. (2024, November 26). Screening for autism spectrum disorder. https://www.cdc.gov/autism/diagnosis/index.htm
Stagg, S. D., & Belcher, H. (2019). Living with autism without knowing: receiving a diagnosis in later life. Health Psychology and Behavioral Medicine, 7(1), 348–361. https://doi.org/10.1080/21642850.2019.1684920
Leedham, A., Thompson, A. R., Smith, R., & Freeth, M. (2020). ‘I was exhausted trying to figure it out’: The experiences of females receiving an autism diagnosis in middle to late adulthood. Autism, 24(1), 135-146.
Machine Learning, Computer Vision,
And Autism
Digital health products will be able to accelerate the diagnostic process, as well as effectively provide information to clinicians conducting assessments.
So far, existing research has uncovered that eye-tracking, combined with machine learning, can be the most effective way
to screen children.
National Autistic Society. (n.d.). Criteria and tools used in an autism assessment. https://www.autism.org.uk/advice-and-guidance/topics/diagnosis/assessment-and-diagnosis/criteria-and-tools-used-in-an-autism-assessment
Keehn, B., Monahan, P., Enneking, B., Ryan, T., Swigonski, N., & McNally Keehn, R. (2024). Eye-Tracking Biomarkers and Autism Diagnosis in Primary Care. JAMA network open, 7(5), e2411190. https://doi.org/10.1001/jamanetworkopen.2024.11190
Liu, X., Zhao, W., Qi, Q., & Luo, X. (2023). A survey on autism care, diagnosis, and intervention based on mobile apps: Focusing on usability and software design. Sensors, 23(14), 6260. https://doi.org/10.3390/s23146260
Ponzo, S., May, M., Tamayo-Elizalde, M., Bailey, K., Shand, A. J., Bamford, R., Multmeier, J., Griessel, I., Szulyovszky, B., Blakey, W., Valentine, S., & Plans, D. (2023). App Characteristics and Accuracy Metrics of Available Digital Biomarkers for Autism: Scoping Review. JMIR mHealth and uHealth, 11, e52377. https://doi.org/10.2196/52377
Hashemi, J., Mahoor, M. H., Esler, A., Sapiro, G., Stone, W., & Campbell, K. (2012). A computer vision approach for the assessment of autism-related behavioral markers. In 2012 IEEE International Conference on Development and Learning and Epigenetic Robotics (ICDL) (pp. 1–7). IEEE. https://doi.org/10.1109/DevLrn.2012.6400865
STAKEHOLDER MAPPING
Through our literature review, we identified autistic individuals and their guardians/parents as the core stakeholders in terms of autistic individuals receiving timely diagnosis and further help.
COMPETITIVE BUSINESS MATRIX
My team and I created a competitive business matrix to see gaps in research and development of services related to autism. Currently, most research projects and applications out in the market target young children, and provide results based on self-reporting. In other words, there is a lack of projects related to teenage autistic individuals that provide an analysis using digital tools.
USER INTERVIEWS
Before designing our application, my team and I were able to interview a total of nine parents and guardians regarding the difficulties related to the diagnostic process. In addition, we have reached out to a total of four clinical experts in autism.
The Diagnostic Process Can Be Intimidating
The Self-Seeking Teenager
주변에서 제가 자폐일 수도 있다고 하더라고요. 제가 어떤 사람인지 제대로 알고 싶긴 하지만, 정말 자폐가 맞을까봐 무섭기도 해서 진단을 받기가 꺼려져요. 또 만약 맞다면 사람들이 저를 다르게 볼까 봐 걱정돼요.
[Translation] People around me have suggested I might be autistic. I do want to understand myself better, but I’m scared it might actually be true, so I’ve been hesitant to get diagnosed. And if it is, I’m worried people might see me differently.
The Wait is Too Long
Some people want to see a diagnostician right away, but they have to wait too long. Some people might not even take the first step because they’re afraid. If there was a simple tool users could just try out easily, it would make things much simpler.
— Expert from Autism Partnership Korea
The wait times at hospitals are extremely long, and it usually takes a year to see a doctor. Even at lesser-known hospitals, it’s still difficult to get an appointment due to long waiting lists. Even if you have the intrinsic motivation to get a diagnosis, it will take a long time for you to actually get it.
— Expert from Korea Psychological Group
HOW MIGHT WE...
Painpoint 1
Users feel frustrated because of the long wait to get a diagnosis.
How Might We...
Help the user get a faster assessment of their current situation.
Painpoint 2
Users feel intimidated about the possibility of having autism.
How Might We...
Help the user feel assured about themselves during the pre-screening and diagnostic process.
Painpoint 3
Users feel confused with where they could find the right resources related to autism.
How Might We...
Help the user find resources (coping methods, clinics, etc.) related to autism more easily.
RESEARCH QUESTION
What if... we make autism screening more accessible for teenage undetected autistic individuals, with AI?”
THE CORE IDEA
A Quick yet Accurate Preliminary Screening Before An Official Diagnosis
KEY FEATURES
Games & Questionnaires
For a quick, easy,
and accurate
pre-screening process
Narrative Intro
& Assuring Reminders
To act as a companion to help users understand who they are
Affiliation with Hospitals & Clinics
To streamline the diagnosis process by providing data that medical experts can use to add to the official diagnosis
CRITICAL USE CASE DIAGRAM
Considering parents and medical experts as key stakeholders in the problem, we have come up with a critical use case diagram featuring potential autistic individuals, medical experts, and parents or guardians.
PROTOTYPING PROCESS
One of the key values my team and I had in mind while creating our application was accessibility. Hence, we were debating where each page should be placed to simplify the process for teenage users, while ensuring the accuracy of the pre-screening.
To design the information architecture of our application, we created a paper prototype to discuss how to keep the users engaged in the testing process.
DESIGNING GAMES FOR PRE-SCREENING
For the games, we decided to include four different types. This is to maximize the accuracy of the application. The four different types are: face scan, quick gaze test, photo description test, and whack-a-mole.
Face Scan (Scanning Facial Landmarks)
Uses the camera to record and analyze facial landmarks of the user.
Research Basis: Based on Perochon et al. (2023) and Drimalla et al. (2020) studies employing the Simulated Interaction Task with automated facial landmark analysis.
Validity Evidence: Validated against expert-labeled emotion data and EMG measures, demonstrating reliable detection of expression differences in autistic versus non-autistic participants.
Accuracy Metrics: Achieved AUC=0.78 with sensitivity=67% and specificity=79%.
Perochon, S., Di Martino, J. M., Carpenter, K. L. H., et al. (2023). Early detection of autism using digital behavioral phenotyping. Nature Medicine, 29(11), 2489–2497. https://doi.org/10.1038/s41591-023-02574-3
Drimalla, H., T. Scheffer, N. Landwehr, et al. 2020. "Towards the Automatic Detection of Social Biomarkers in Autism Spectrum Disorder: Introducing the Simulated Interaction Task (SIT)." NPJ Digital Medicine 3: 25. https://doi.org/10.1038/s41746-020-0227-5.
Quick Gaze Test (Gaze Tracking)
Tracks the gaze of users and checks patterns related to autis.
Research Basis: Based on Pierce et al. (2023) “SenseToKnow” study and computer vision methods of Hashemi et al. (2012) for assessing autism-related gaze markers.
Validity Evidence: Validated through comparison with clinical diagnoses and cross-study replication in both clinical and home settings; Hashemi et al. demonstrated reliable gaze-pattern extraction against expert behavioral ratings.
Accuracy Metrics:
-
Combined digital phenotype: AUC = 0.90, sensitivity = 87.8%, specificity = 80.8%
-
Hashemi et al.: reported over 85% agreement with expert annotations for autism-related gaze behaviors.
Pierce, K., Boudreau, L., Joyce, J., et al. (2023). Early detection of autism using digital behavioral phenotyping. Nature Medicine, 29, 1929–1936. https://doi.org/10.1038/s41591-023-02574-3
Hashemi, J., Mahoor, M. H., Esler, A., Sapiro, G., Stone, W., & Campbell, K. (2012). A computer vision approach for the assessment of autism-related behavioral markers. In 2012 IEEE International Conference on Development and Learning and Epigenetic Robotics (ICDL) (pp. 1–7). IEEE. https://doi.org/10.1109/DevLrn.2012.6400865
Video Description Test (Speech Feature Analysis)
Records the user describing short video clips (baseline test, social situation, non-social situation), analyzes their speech features, and compares attention to three different types of situations.
Research Basis: Based on Briend et al. (2023) and similar voice-acoustic classification studies using nonword and video-description tasks, and Egger et al. (2018) feasibility study of automatic emotion and attention analysis in young children at home using ResearchKit.
Validity Evidence: Validated using cross-validation, ROC analysis, and comparison against typically developing and mixed-control groups, alongside home-based validation of automated emotion/attention detection.
Accuracy Metrics: Classification accuracy of 91% against typically developing children and 85% against a mixed control group.
Briend, F., Silleresi, S., David, C., Bonnet-Brilhault, F., Petit, L., & Dagneau, M. (2023). Voice acoustics allow classifying autism spectrum disorder with high accuracy. Translational Psychiatry, 13, Article 245. https://doi.org/10.1038/s41398-023-02554-8
Egger, H. L., Dawson, G., Hashemi, J., Carpenter, K. L. H., Espinosa, S., Campbell, K., Brotkin, S., Shaich-Borg, J., Qiu, Q., Tepper, M., Baker, J. P., Bloomfield, R., & Sapiro, G. (2018). Automatic emotion and attention analysis of young children at home: A ResearchKit autism feasibility study. npj Digital Medicine, 1(1), Article 20. https://doi.org/10.1038/s41746-018-0024-6
The Whack-A-Mole (Motor Assessment)
A game where users tap moving targets on the screen, measuring reaction speed, touch accuracy, and movement patterns.
Research Basis: Based on Perochon et al. (2023) study of motor interaction modules within tablet-based autism screening (“SenseToKnow” motor task) and related gamified motor assessments.
Validity Evidence: Validated against clinical diagnoses and shown to capture distinctive motor control patterns in autistic children through cross-study replication and expert review.
Accuracy Metrics: Contributed to combined digital phenotype AUC=0.90, sensitivity=87.8%, specificity=80.8% when motor data were integrated with gaze and other biomarkers.
Perochon, S., Di Martino, J. M., Carpenter, K. L. H., et al. (2023). Early detection of autism using digital behavioral phenotyping. Nature Medicine, 29(11), 2489–2497. https://doi.org/10.1038/s41591-023-02574-3
LO-FI PROTOTYPE
3rd Iteration
A Narrative Style Intro
Self-Questionnaire
Homepage
Games for
Pre-screening
Results
Usability Testing
After creating storyboards according to each use case (teenagers, parents, experts), we interviewed two non-autistic teens, one autistic teen, one parent of an autistic teen, and two experts to gather feedback on our prototype.
USER FEEDBACK
We were able to meet with a total of six participants to gather user feedback.
91%
of participants said they would want to download
and use this app.
91%
of participants reported that
the narrative intro and reminders helped them feel assured about their self-identity.
83%
of participants agreed that the app
can help them find resources
for autistic individuals.
Teenage Users
Non-autistic Teen: (KOR → ENG Translated)
"I would be willing to use this application if I ever suspected myself to be autistic."
Autistic Teen: (KOR → ENG Translated)
"I think giving out reminders would be effective in reducing the anxieties users may have."
Suggestions:
-
Add an explanation of why the facial scan is being done to reduce privacy concerns
INFORMATION ARCHITECTURE
Below is the final information architecture of the application that
we have come up with after discussing with our paper prototype.
Narrative Intro
Take Tests (Games)
Answer Self-Questionnaires (AQ-12)*
Create Account
View Results
Homepage
Check Results Again
Discover Next Steps
Discover Coping Methods for Autism (Tips and Tricks)
Discover Associations
& Groups
Find Affiliated
Clinics Nearby
Try Tests (Games) Again
Try Tests (Games) Again
Change Settings
*AQ-12: The AQ-12 questionnaire is a short form of the Autism-Spectrum Quotient (AQ), designed to efficiently screen for traits associated with autism spectrum disorder using just 12 items instead of the original 50-item AQ. According to Lundqvist (2017), the AQ-12 was developed through Rasch analysis of the full AQ, and this abbreviated version retains strong explanatory power for quantifying autistic traits in adults with normal intelligence.
Lundqvist, L.-O., & Lindner, H. (2017). Is the Autism-Spectrum Quotient a Valid Measure of Traits Associated with the Autism Spectrum? A Rasch Validation in Adults with and Without Autism Spectrum Disorders. Journal of Autism and Developmental Disorders, 47(7), 2080–2091. https://doi.org/10.1007/s10803-017-3128-y
Hoekstra, R. A., Vinkhuyzen, A. A. E., Wheelwright, S., Bartels, M., Boomsma, D. I., Baron-Cohen, S., Posthuma, D., & van der Sluis, S. (2011). The construction and validation of an abridged version of the autism-spectrum quotient (AQ-Short). Journal of Autism and Developmental Disorders, 41(5), 589-596.
Allison, C., Auyeung, B., & Baron-Cohen, S. (2012). Toward brief "red flags" for autism screening: The short autism spectrum quotient and the short quantitative checklist in 1,000 cases and 3,000 controls. Journal of the American Academy of Child & Adolescent Psychiatry, 51(2), 202-212.
Broadbent, J., Galic, I., & Stokes, M. A. (2013). Validation of autism spectrum quotient adult version in an Australian sample. Autism Research and Treatment.
Kuenssberg, R., Murray, A. L., Booth, T., & McKenzie, K. (2014). Structural validation of the abridged Autism Spectrum Quotient–Short Form in a clinical sample of adults with autism spectrum disorders. Autism, 18(1), 69-75.
ADDITION OF NEW PAGES
The following are key enhancements added after our usability testing sessions.
Providing Assurance Before Each Test
Our tests showed users liked friendly, non-intrusive nudges. Adding on to the idea of "a companion throughout your journey of self-discovery," the custom reminders such as “Don't worry, I'll be there with you” encourage users to keep going without feeling pressured or anxious.
FINAL PRODUCT
Every autistic individual deserves timely understanding
Understanding Yourself
Starts Here
-
Our mascot, Miso, will always be there to guide you through your journey of discovering who you are.
-
Take the first step toward understanding your behaviors and traits with curiosity and confidence.
Fun that reveals,
not judges
-
Explore who you are through gamified tests and intuitive self-questionnaires.
-
Each game is designed to capture meaningful behavioral data, instead of labeling or evaluating.
Awareness Starts Acceptance
-
Take our guide through personalized insights and educational resources.
-
Explore data that helps you know yourself, not define.
Take gentle steps
at a time
-
Read tips and tricks for coping with autism, find groups and associations, or find autism clinics nearby to start your next step.
-
Find strategies, stories, and spaces that help you thrive. From curiosity to care, move forward at your own comfort level.
Impact
This research directly tackles the significant problem of diagnostic delays. We found that families typically wait about a year for an autism diagnosis. Furthermore, using a multi-modal biomarker approach with AI detection tools, our study contributes to this growing body of evidence supporting digital screening approaches.
Overall, while designing our prototype, we aimed to take a user-centered design approach with extensive stakeholder engagement (including teenagers, parents, and medical experts) to provide a robust framework in developing accessible digital health tools for teenage autistic individuals.
Limitations and Future Research Directions
-
Limited device compatibility (only smartphone prototype)
-
Need for larger-scale validation studies
My Key Takeaways
-
Be mindful of how users feel during each medical process: At first, I was more focused on what kind of new technologies we could use to create a pre-screening service. However, when I got to meet actual autistic individuals, guardians, and clinical experts, I realized that providing mental support during the diagnostic process is also important for our potential users.
-
Keep in mind of the prototype's goal: With our first design of the application's flow, after viewing the results page, users would be directly linked to the page that helps them find clinics nearby. However, after conducting usability testing sessions, we realized that linking them right away to the clinics page may frighten the users about their situation, which does not fit our application's goal of providing a less intimidating pre-screening to users. Hence, to match our goal, we decided to add in a page that gives tips on how to cope with their situation in their daily lives, as well as a page that introduces them to groups and associations where they could receive help.
-
Understand how different populations interact with technology: We tried to incorporate specific design considerations for autistic users (sans serif fonts, bright colors, simple navigation, calming mascot design). Here, I learned that accessibility in digital health extends beyond compliance.