GAN-Based Privacy-Preserving Intelligent Medical Consultation Decision-Making

Abstract

In the era of big data, information leakage during medical consultation has become a crucial factor in patients’ decision-making. This paper presents an intelligent medical decision model that considers patient privacy. The model utilizes data synthesized through a generative adversarial network (GAN) for intelligent training, ensuring privacy protection. First, we formulate a risk-based decision model for three different alternative medical consultation modes, analyzing decision rules related to visiting distance and disease probability. Next, we construct a data-driven intelligent medical decision framework. To address privacy concerns, we employ GAN to generate synthetic data from historical patient records, which is seamlessly incorporated into the decision framework to derive decision rules. Finally, specific patient data is utilized to make informed medical decisions. We validated our model using the random forest algorithm and liver disease patients’ medical decisions. Empirical findings demonstrate that the GAN-based synthetic data improves the nearest-neighbor distance ratio by 12.4% compared to synthetic data with Gaussian noise, thereby enhancing data privacy. Additionally, the GAN-based prediction models outperform the models trained on real data, achieving a notable increase of 6.3% and 4.1% in average accuracy and F1 score, respectively. Notably, the GAN-based intelligent decision-making models surpass four other baseline medical visit decision-making methods with an impressive accuracy of 74.0%. In conclusion, our proposed intelligent medical decision-making model effectively prioritizes user data privacy while enhancing the quality of medical decision-making.