Crowd annotations can approximate clinical autism impressions from short home videos with privacy protections.

Artificial Intelligence (A.I.) solutions are increasingly considered for telemedicine. For these methods to serve children and their families in home settings, it is crucial to ensure the privacy of the child and parent or caregiver. To address this challenge, we explore the potential for global image transformations to provide privacy while preserving the quality of behavioral annotations. Crowd workers have previously been shown to reliably annotate behavioral features in unstructured home videos, allowing machine learning classifiers to detect autism using the annotations as input. We evaluate this method with videos altered via pixelation, dense optical flow, and Gaussian blurring. On a balanced test set of 30 videos of children with autism and 30 neurotypical controls, we find that the visual privacy alterations do not drastically alter any individual behavioral annotation at the item level. The AUROC on the evaluation set was 90.0% ±7.5% for unaltered videos, 85.0% ±9.0% for pixelation, 85.0% ±9.0% for optical flow, and 83.3% ±9.3% for blurring, demonstrating that an aggregation of small changes across behavioral questions can collectively result in increased misdiagnosis rates. We also compare crowd answers against clinicians who provided the same annotations for the same videos as crowd workers, and we find that clinicians have higher sensitivity in their recognition of autism-related symptoms. We also find that there is a linear correlation (r = 0.75, p < 0.0001) between the mean Clinical Global Impression (CGI) score provided by professional clinicians and the corresponding score emitted by a previously validated autism classifier with crowd inputs, indicating that the classifier's output probability is a reliable estimate of the clinical impression of autism. A significant correlation is maintained with privacy alterations, indicating that crowd annotations can approximate clinician-provided autism impression from home videos in a privacy-preserved manner.

Training Affective Computer Vision Models by Crowdsourcing Soft-Target Labels.

Background/Introduction: Emotion detection classifiers traditionally predict discrete emotions. However, emotion expressions are often subjective, thus requiring a method to handle compound and ambiguous labels. We explore the feasibility of using crowdsourcing to acquire reliable soft-target labels and evaluate an emotion detection classifier trained with these labels. We hypothesize that training with labels that are representative of the diversity of human interpretation of an image will result in predictions that are similarly representative on a disjoint test set. We also hypothesize that crowdsourcing can generate distributions which mirror those generated in a lab setting. Methods: We center our study on the Child Affective Facial Expression (CAFE) dataset, a gold standard collection of images depicting pediatric facial expressions along with 100 human labels per image. To test the feasibility of crowdsourcing to generate these labels, we used Microworkers to acquire labels for 207 CAFE images. We evaluate both unfiltered workers as well as workers selected through a short crowd filtration process. We then train two versions of a ResNet-152 neural network on soft-target CAFE labels using the original 100 annotations provided with the dataset: (1) a classifier trained with traditional one-hot encoded labels, and (2) a classifier trained with vector labels representing the distribution of CAFE annotator responses. We compare the resulting softmax output distributions of the two classifiers with a 2-sample independent t-test of L1 distances between the classifier's output probability distribution and the distribution of human labels. Results: While agreement with CAFE is weak for unfiltered crowd workers, the filtered crowd agree with the CAFE labels 100% of the time for happy, neutral, sad and "fear + surprise", and 88.8% for "anger + disgust". While the F1-score for a one-hot encoded classifier is much higher (94.33% vs. 78.68%) with respect to the ground truth CAFE labels, the output probability vector of the crowd-trained classifier more closely resembles the distribution of human labels (t=3.2827, p=0.0014). Conclusions: For many applications of affective computing, reporting an emotion probability distribution that accounts for the subjectivity of human interpretation can be more useful than an absolute label. Crowdsourcing, including a sufficient filtering mechanism for selecting reliable crowd workers, is a feasible solution for acquiring soft-target labels.