UF Researchers are working to anonymize video data in autism spectrum disorder studies.
Researchers at the University of Florida are tackling a significant challenge in the diagnosis and treatment of autism spectrum disorder (ASD) in children. The challenge is clinician availability, which limits access to diagnosis and care. However, advances in computational methods and crowdsourcing hold promise for expanding the availability of treatment, but to achieve this, researchers need access to video data on children with ASD. When working with video data, patient privacy becomes a key concern.
Ensuring the confidentiality of patient data is crucial when sharing video recordings that contain sensitive information, such as the patient’s face and voice. The standard approach to anonymization involves removing or modifying personal identifiers, which is easily achieved with text data. However, with video data, the challenge is more complex. Anonymizing facial and vocal identifiers without altering the critical details needed to understand ASD-specific behaviors has traditionally been an issue.
To overcome this challenge, Eakta Jain, Ph.D., principal investigator and an associate professor at the UF Department of Computer & Information Science & Engineering, and Kevin Butler, co-investigator and a professor at the UF Department of Computer & Information Science & Engineering, have received a $3.7 million grant from the National Institutes of Health (NIH) to develop innovative solutions. Jain plans to use the latest in AI based models to modify the faces and voices of patients while retaining the essential information needed to quantify ASD-associated behaviors. This approach aims to create a privatized data set of video recordings that balances patient confidentiality with researcher access.
According to Jain, existing algorithms for anonymizing audio and video data, such as blur filters, are not effective for retaining gaze and facial expression, which are critical details when clinicians screen for autism.
“Though there is a large body of work in anonymization in both audio and video processing, it is yet unknown how well existing algorithms obfuscate identity while retaining autism-specific atypicality,” Jain explained.
With a privatized data set, Jain hopes to facilitate future research and provide a valuable resource for expanding access to data resources for clinician education and computational research. This could ultimately lead to improved diagnosis and care for children with ASD, making treatment more accessible and effective.
By Drew Brown
Marketing and Communications Specialist