Figure 1 | Scientific Reports

Figure 1

From: Crowdsourced privacy-preserved feature tagging of short home videos for machine learning ASD detection

Figure 1

Overview of the crowd-powered AI detection process. (a) A trustworthy crowd is selected through a filtration process involving an evaluation set of videos. (b) A diagnosis and gender balanced set of unstructured videos are evaluated both with and without a set of privacy-preserving alterations: pitch shift and face obfuscation. (c) The curated crowd extracts behavioral features about the children in the videos by answering a set of multiple choice questions about the child’s behavior exhibited in the video, with each worker assigned to a random subset of the videos. (d) A classifier trained on electronic medical records (the “training set”) corresponding to the multiple choice answers to behavioral questions is used to predict the diagnosis from the aggregated video-wide annotations (the “test set”), and the classifications are compared against the known diagnoses in the video set (the “test set”).

Back to article page