An Efficient and Zero-Trust Approach for Privacy-Preserving Healthcare Diagnostics

The integration of the Internet of Things (IoT) into the healthcare industry has led to the development of the Internet of Medical Things (IoMT). In IoMT, healthcare professionals diagnose and treat patients by analyzing data collected from medical sensors, with measurements transmitted via AI-powered mobile applications. These health records are subsequently processed through machine learning (ML) models for disease diagnosis <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$(\mathcal{DD})$</tex> and prediction. However, exposing such sensitive data raises privacy concerns, as it may lead to the inference of confidential patient information. To mitigate this issue, existing research primarily employs federated learning (FL)-based strategies to collaboratively train and generate an accurate global <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathcal{DD}$</tex> model across multiple healthcare institutions. Despite these efforts, a significant gap remains in addressing privacy risks during the future <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathcal{DD}$</tex> process in the deployment phase, once the global model has been established-an aspect that has not been thoroughly explored. In response, this paper introduces a novel decentralized privacy-preserving scheme for accurate <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathcal{DD}$</tex> while ensuring patient data confidentiality. The proposed approach enables patients to encrypt their medical records using functional encryption (FE) without relying on a trusted key distribution center (KDC), thereby allowing <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathcal{DD}$</tex> without exposing their health data. Additionally, we design a hybrid deep learning (DL) model to enhance diagnostic accuracy. To validate our approach, we evaluate its performance on real-world health records from the Cleveland dataset, sourced from the University of California Irvine (UCI). Our results demonstrate that the proposed scheme effectively diagnoses heart disease while maintaining robustness, preserving privacy, and minimizing computational overhead.