Safely Using Image Data to Improve AI Algorithms

Safely Using Image Data to
Improve AI Algorithms

Client Context

Company Uses X-Ray Data for Software R&D

A medical device company produces software that uses AI to derive diagnoses from medical images. The medical image data comprises X-ray image files.  These files include both pixel data visualizing certain anatomy and a variety of metadata tags stored in the DICOM header.

The images are sourced from hospital partners and shared internally with an R&D team to improve the AI algorithm. This use of data is considered “secondary” under HIPAA, and therefore requires the data to be non-identifiable (de-identified).  To de-identify the data while also achieving the data utility needed by the company, an expert determination approach under HIPAA was required.

Business Problem

Company Needs to Validate Existing De-Identification Process

This company had developed a de-identification strategy that balanced their need for high-quality data with the regulatory requirement that the data be non-identifiable. However, before expanding their program, they wanted to be completely sure their strategy was fully compliant with HIPAA.


Privacy Analytics Assesses the Effectiveness of the De-identification Strategy

The company’s existing de-identification strategy included various techniques to protect individual privacy, including masking and generalization.  Privacy Analytics measured the identifiability (or re-identification risk) after the customer applied their strategy to the data.

In addition to evaluating the data itself (i.e., how distinguishable individuals are in the data), we also considered the context of the intended data use, including security controls.

In our initial assessment, we found that the data was still identifiable and recommended a strategy to further de-identify data while preserving its value. With this refined strategy, our customer is now confident in their approach and can achieve their desired data utility.


Improved AI Algorithms Thanks to Full Dataset Confidence

With improved AI algorithms, the customer’s products can generate more accurate diagnoses.  Higher accuracy leads to improved patient outcomes, allowing for stronger results-driven product claims and differentiation – a must in today’s hyper-competitive device market.

Learn more about Privacy Analytics’ anonymization services for DICOM.

More Case Studies


Join the next 5 Safes Data Privacy webinar

This course runs on the 2nd Wednesday of every month, at 11 a.m. ET (45 mins). Click the button to register and select the date that works best for you.