BNN-based Privacy-aware Healthcare Data Analysis

Recent advances in Neural Networks have shown increasing success in analyzing healthcare data. But they also raise challenges to efficient computation and data privacy. This paper proposes a novel privacy preserving healthcare data analytics framework that, while ensuring privacy of medical data, supports efficient data analysis and requires fewer parameters by introducing Binary Neural Networks (BNN). Experimental results over dataset MedMNIST indicate that the proposed framework exhibits favorable performance across multiple healthcare datasets. The BNN model achieves high accuracy while retaining a minimal number of parameters. Furthermore, regarding privacy inference performance, it demonstrates shorter inference times and higher accuracy in comparison to current mainstream networks such as ResNet-20.