Efficient Online and Privacy-preserving Medical Pre-diagnosis Based on Growing Learning Vector Quantization

Abstract

With more and more people turning to online medical pre-diagnosis systems, it becomes increasingly important to protect patient privacy and enhance the accuracy and efficiency of diagnosis. That is because the ever rapidly growing medical records not only contain a large amount of private information but are often highly unequally distributed (e.g., the number of cases and the rate of increase of covid-19 can be much higher than that of common diseases). However, existing methods are not capable of simultaneously boosting the intensity of privacy protection, and the accuracy and efficiency of diagnosis. In this paper, we propose an online medical pre-diagnosis scheme based on incremental learning vector quantization (called WL-OMPD) to achieve the two objectives at the same time. Specifically, within WL-OMPD, we design an efficient algorithm, Wasserstein-Learning Vector Quantization (W-LVQ), to smartly compress the original medical records into hypothetic samples. Then, we transmit these compressed data to the cloud instead of the original records to offer a more accurate pre-diagnosis. Extensive evaluations of real medical datasets show that the WL-OMPD scheme can improve the imbalance ratio of the data to a certain extent and then the intensity of privacy protection. These results also demonstrate that WL-OMPD substantially boost the accuracy of the classification model and increase diagnostic efficiency at a lower compression rate.