Artificial Intelligence

Amazon is turning its controversial Rekognition software to medical image redaction

Amazon has demonstrated the ability to combine its Comprehend Medical software service with its Rekognition image analysis to help automate medical image de-identification.

Six panels of an MRI scan.

As part of the tech giant’s efforts in healthcare, Amazon is pitching its Amazon Rekognition image and video analysis service as a way more effectively de-identify and utilize medical images.

The digitization of medical images and the introduction of new types of imaging technology has vastly increased the volume and utility of medical images as part of research and clinical practice.

Medical image archives continue to grow – along with new AI-based techniques to analyze them – but the often manual redaction and protection of PHI on these images is a challenge in bringing these new technologies fully to bear.

Amazon has demonstrated the ability to combine its Amazon Comprehend Medical software service, which uses machine learning and natural language processing to extract relevant clinical information from records, with its Rekognition image analysis to help automate this de-identification process. Both software services are offered through the Amazon Web Services cloud computing platform.

Essentially, Rekognition is able to scan medical images for text and Comprehend Medical can understand which of that data is PHI and needs to be removed or redacted creating a large usable dataset while ostensibly protecting patient privacy.

One important note of caution from the company is that service (like any other AI-based technology) is not perfect, instead providing a confidence score describing the accuracy of the detected information.

Rekognition was initially launched in 2016 as a developer tool focused on video and image analysis. Over the years, the service has been bolstered with a wide range of applications ranging from tracking objects in videos and making it easier to categorize image and video archives.

Where the tool has seen some controversy is over its facial recognition capabilities, which has used by law enforcement agencies to identify criminals or suspects. The company also reportedly pitched the product for use by Immigration and Customs Enforcement, which led to complaints about potential abuses of the technology.

By and large though, the enterprise uses of Rekognition have been more innocuous. Some of the case studies published by Amazon include companies like Scripps Networks Interactive, which uses the tool for automated metadata tagging which makes it easier to search for and find specific images or videos.

Photo: maciu17, Getty Images