Adversarial learning for De-Identification of Medical-Records

Medical records may exclusively be disclosed for the purpose of research. Medical researchers can only access the medical records of patient if they are de-identified because of data protection and privacy laws. De-Identified data means the process of removing or masking protected health information (PHI) in order to reduce the risk that subjects identify be connected with data. Automatic de-identification classifiers The goal is to make it affordable to de-identify large amounts of medical records. Current classifiers are trained on costly manually pseudonymized records from a single source, limiting their applicability to new data sources. To improve this, we need larger, more diverse datasets for training more versatile de-identification tools. We propose a method to convert medical text into a secure format without the need for pseudonymization, making it easier to share training data securely.