Computer Vision to Monitor Sleep Study Participants

The Problem:

Our client was a pharmaceutical company who was conducting sleep studies for patient suffering from severe dermatitis. During the course of the sleep studies, the patients would experience many scratching events. These sleep studies were recorded with a camera, and the results were quantified manually by reviewing the footage afterward. This process was extremely time and labor intensive.

The Data:

The client provided camera footage from all sleep study participants, which consistent of hundreds of hours per participant. 

The Solution:

We created a custom software tool to allow initial reviewers view footage and identify scratched scratch events, then log the event with the correct file and timestamp. Using the labels generated by footage reviewers, we trained a computer vision algorithm to detect scratching events automatically for new footage. We deployed the model in the client's cloud environment.

The Impact:

The client was able to use our model to automatically detect and count scratching events in new footage. While there was still some manual effort to spot-check the results, the overall manual labor involved in the process was reduced by over 80%.