Thanks to improvements in digital imaging and data storage, medical archives are much larger than they’ve ever been. Researchers now have access to millions of MRIs, X-Rays, and other images that they can analyze to better understand, diagnose, and treat any number of medical conditions.
The catch is that doctors need to be able to guarantee the privacy of their patients, making those images useless if they contain any privileged personal information. Scanning and redacting each individual image can be prohibitively time consuming and expensive, especially when dealing with a larger dataset.
That’s why Amazon is pitching Amazon Rekognition and Amazon Comprehend Medical as a two-step solution that can automate the de-identification process. Amazon Web Services demonstrated the technique in a recent how-to post on the AWS Machine Learning Blog. It requires a bit of programming, but the concept itself is relatively simple: Rekognition is able to extract any text embedded in a medical image. Comprehend Medical then scans that text and redacts any protected health information.
Together, the two platforms are able to generate a large, usable body of data that also guarantees clinical privacy. The method isn’t foolproof, but it does produce a confidence score so doctors can go through and verify questionable examples.
If Amazon Rekognition does get used to facilitate medical research, it would likely be one of the more socially conscious applications of the technology. Rekognition has repeatedly come under fire from various civil rights groups, although Amazon has been trying to highlight some of the non-surveillance uses to defended the platform. The recent blog post is in keeping with that trend, and indicates that Amazon is still invested in the utility of Rekognition.
March 21, 2019 – by Eric Weiss