EdgeCare: privacy-preserving medical advising system on mobile devices

Abstract

Ensuring the confidentiality of information and accuracy especially related to medical data is a critical challenge in the development of digital health applications. This paper presents a novel approach for a medical chat application that is intended to preserve user privacy while ensuring the accuracy of responses. On-device privacy-preserving techniques and context aware medical report retrieval mechanisms are engaged on Android mobile phones with cloud-based retrieval-augmented generation (RAG) in this system. A lightweight, transformer based language model is leveraged for the anonymization of protected health information (PHI) directly on the user’s mobile device with a medical report storage and a retriever ensuring private and sensitive information never leaves the device in its raw form. The cloud-based subsystem acts as the backend and is responsible for processing the anonymized requests, retrieving relevant medical knowledge, and generating accurate, context aware responses using a large language model (LLM).