Research on the Medical Structured Data Desensitization Method Based on a Bayesian Network

To address the prevalent privacy concerns associated with patient data in current clinical medical data analysis, this paper proposes a medical structured data desensitization method based on Bayesian Network. The proposed method involves constructing a data generation model by learning the statistical and correlational attributes of the original dataset, thereby producing a desensitized dataset that maintains characteristic probability similarity with the source data. The experimental results, based on the analysis of clinical data, demonstrate that the desensitized data generated by this method exhibits exceptional performance in preserving intrinsic data feature correlations while maximizing data utility. In machine learning tasks, the desensitized data can achieve comparable model performance indicators to those obtained from the original data. The chi-square (x2) test results indicate a significant consistency in probability distributions between the desensitized data and the original data, demonstrating that the desensitized data serves as a reliable substitute for the original data in statistical analyses. The proposed method not only provides a practical solution for clinical data desensitization but also offers a novel pathway to overcome data silo challenges in medical artificial intelligence applications, thereby facilitating secure data sharing and collaboration in healthcare research.