Title:An Optimized Information-Preserving Relational Database Watermarking Scheme for Ownership Protection of Medical Data

Abstract:Recently, a significant amount of interest has been developed in motivating physicians to use e-health technology (especially Electronic Medical Records (EMR) systems). An important utility of such EMR systems is: a next generation of Clinical Decision Support Systems (CDSS) will extract knowledge from these electronic medical records to enable physicians to do accurate and effective diagnosis. It is anticipated that in future such medical records will be shared through cloud among different physicians to improve the quality of health care. Therefore, right protection of medical records is important to protect their ownership once they are shared with third parties. Watermarking is a proven well known technique to achieve this objective. The challenges associated with watermarking of EMR systems are: (1) some fields in EMR are more relevant in the diagnosis process; as a result, small variations in them could change the diagnosis, and (2) a misdiagnosis might not only result in a life threatening scenario but also might lead to significant costs of the treatment for the patients. The major contribution of this paper is an information-preserving watermarking scheme to address the above-mentioned challenges. We model the watermarking process as a constrained optimization problem. We demonstrate, through experiments, that our scheme not only preserves the diagnosis accuracy but is also resilient to well known attacks for corrupting the watermark. Last but not least, we also compare our scheme with a well known threshold-based scheme to evaluate relative merits of a classifier. Our pilot studies reveal that -- using proposed information-preserving scheme -- the overall classification accuracy is never degraded by more than 1%. In comparison, the diagnosis accuracy, using the threshold-based technique, is degraded by more than 18% in a worst case scenario.