De-Identification of Histopathology Slide Images, the NHLBI BioData Catalyst® Data Management Core Experience

Abstract

Histopathology whole slide images (WSIs) are increasingly incorporated into the NIH/NHLBI BioData Catalyst (BDC) ecosystem to support secondary research and large scale analytics. While WSIs enable rich, multi resolution image analysis, their complex file structure and proprietary formats (such as SVS) pose challenges for de identification. This presentation provides a focused overview of the WSIs in the SVS format, including embedded metadata, associated images (label and macro), and tiled image pyramids, and identifies common locations where personally identifiable information (PII) may persist.

We demonstrate methods for inspecting SVS contents and present a pipeline for PII detection that combines metadata parsing, NLP based entity recognition, and OCR based image scanning. We discuss technical limitations related to image scale, tiling, and memory constraints, as well as the risks of file corruption during modification. We conclude with an evaluation of available de identification tools and some practical recommendations for data curators and contributors.