Open video data sharing in developmental science and clinical practice.

Area of Science:

• Behavioral Science
• Neuroscience
• Computer Science

Background:

• Sharing large-scale video datasets in behavioral research is crucial but hindered by confidentiality concerns.
• Data-heavy computational methods exacerbate the need for pooled datasets while increasing privacy risks.

Purpose of the Study:

• To investigate if data de-identification techniques compromise the utility of video data for neurological assessments.
• To evaluate face-blurring as a viable method for de-identifying behavioral videos while preserving diagnostic accuracy.

Main Methods:

• Utilized a video-based diagnostic tool for assessing neurological deficits.
• Applied face-blurring (pseudonymization) to video recordings of infant neuromotor functions.
• Compared classification accuracy of de-identified videos by human assessors and artificial intelligence (AI) methods.

Main Results:

• Face-blurring did not significantly impact the classification accuracy for diagnosing infant neuromotor functions.
• Both human experts and AI algorithms maintained high accuracy when analyzing de-identified video data.
• Pseudonymization via face-blurring proved to be an adequate and easily implementable solution.

Conclusions:

• Face-blurring is a viable strategy for de-identifying behavioral video data, ensuring privacy protection.
• This approach facilitates the sharing and pooling of video datasets, crucial for advancing computational behavioral research.
• Encourages the development of innovative solutions for secure data sharing to benefit scientific and public health initiatives.