Medical imaging privacy: A systematic scoping review of key parameters in dataset construction and data protection

Abstract

Background

With digitalization in the field of healthcare, using patient image based data, there is also increasing concerns on protection of patient privacy. Globally various legal rules and regulations have been adopted for stringent measures on data privacy. However, despite the growing importance of privacy, there are currently no universally defined protocols outlining the specific parameters for the de-identification/pseudo-anonymization of medical images.

Objectives

The study aims to assess current methods for protecting patient privacy in medical image datasets used in research and healthcare technology development.

Methods

A comprehensive, systematic search was conducted with a defined search string across databases, including PubMed/Medline, Scopus, Web of Science, Embase, and Google Scholar. Studies were selected based on their focus on the procedures used for anonymization, pseudo-anonymization, and de-identification of medical images during the creation of datasets.

Results

From an initial pool of 324 potentially relevant articles, 13 studies were ultimately included in the final review after meeting the inclusion criteria. Of these, the majority focused on open-source datasets, which are accessible for use in research and algorithm development. Methods of de-identification of images included burn-in annotation, defacing processes, removal of DICOM tags, and facial de-identification. A medical image protection checklist was created based on the findings of our review.

Discussion

The review explores techniques such as removal or masking of personal identifiers, DICOM tag removal, facial de-identification

Goal

The insights gathered aim to help develop standardized privacy protocols to be adhered by healthcare professionals for responsible use of medical imaging data, ensuring the responsible use of medical imaging data for healthcare advancements.

Conclusion

The findings of this review highlight several key considerations for effective pseudo-anonymization and de-identification of medical images. The review emphasizes the need for a careful balance between protecting patient privacy and ensuring that medical datasets retain sufficient quality and richness for research and technological development.