ARTPHIL: Reversible De-identification of Free Text Using an Integrated Model

Abstract

AbstractOrganisations that collect and maintain individual data face the challenge of preserving privacy and security when using, archiving, or sharing these data. De-identification tools are essential for minimising the privacy risk. However, current data de-identification and anonymisation methods are widely used to alter the original data in a way that cannot be recovered. This results in data distortion and, hence, the substantial loss of knowledge within the data.To address this issue, this paper introduces the concept of reversible data de-identification methods to de-identify unstructured health data under the Health Insurance Portability and Accountability Act (HIPAA) guidelines. The model integrates Philter [9], the state-of-the-art tool for extracting personal identifiers from free-text, to detect confidential information and encrypt them with E-ART, lightweight encryption algorithm E-ART [10]. The performance of the proposed model ARTPHIL is evaluated using i2b2 data corpus in terms of recall, precision, F-measure and execution time. The results of the experiment are consistent with the recent de-identification method with recall of 96.93%. More importantly, the original data can be recovered, if needed, and authenticated.KeywordsPrivacyDe-identificationPseudonymisationReversibleRe-identification