Privacy-preserving computation and robust watermarking techniques for healthcare data
Item Usage Stats
MetadataShow full item record
Please cite this item using this persistent URLhttp://hdl.handle.net/11693/47725
Health and genomic data is sensitive in terms of carrying private information about individuals. One can infer inherited/genetic disorders, their occurrence probabilities, information about race, and kinship by analyzing an individual's genomic data. Furthermore, health data which is mostly collected by hospitals or other health institutions carries private information about individuals including the diseases they have at present or indicators of future diseases/disorders. While protecting such data, it is important to show that its utility is preserved and maximized since the data is used in researches. Regarding these facts, homomorphic encryption-based scheme (using Paillier cryptosystem) for the protection of health data and a novel watermarking scheme based on belief propagation algorithm for the genomic data is proposed in this work. Homomorphic encryption is used for the health data to show the ability of performing mathematical operations on the encrypted data without decrypting it with a real-life use-case. We show its practicality with the correctness and performance results. In the second part of this thesis, a watermarking scheme for genomic data is proposed to overcome the liability issues due to unauthorized sharing by service providers (SPs). Robust-watermarking techniques ensure the detection of malicious parties with a high probability and we show the probabilistic limits of this detection with di erent experimental setups and evaluation metrics. Lastly, this scheme guarantees the following with a high probability: (i) the utility is preserved, (ii) it is robust against single or colluding SP attacks, and (iii) watermark addition is compatible with the nature of the data as the proposed method considers auxiliary information that a malicious SP may use in order to remove/modify watermarked points before leaking the data.