KArAt-KidneyNet: Proactive Threat Hunting and Anomaly Detection Framework for Consumer Wearable Healthcare Electronics in Kidney Disease Diagnosis

Document Type

Article

Publication Date

1-1-2026

Abstract

Consumer wearable healthcare electronics, including kidney-monitoring wearables and point-of-care diagnostic devices, generate continuous multimodal physiological and imaging data. Their increasing integration within connected health systems also expands the attack surface, creating vulnerabilities that may affect diagnostic integrity, model reliability, and patient safety. This paper presents KArAt-KidneyNet, a unified framework for kidney disease diagnosis and system-state monitoring that combines Kolmogorov-Arnold Transformer (KArAt) attention, Soft Actor-Critic (SAC)-based adaptive defense, and privacy-preserving federated learning. The proposed KArAt attention module replaces softmax attention with a set of learnable basis functions inspired by the Kolmogorov-Arnold representation concept, enabling structured feature decomposition and interpretable attention maps for ultrasound imaging. The SAC component provides adaptive parameter adjustment for anomaly-aware defensive actions based on system state indicators derived from detection modules. Federated learning with differential privacy is used to train the diagnostic model over decentralized healthcare datasets without sharing raw patient data. Additional protective layers include post-quantum cryp-tographic signatures resistant to emerging quantum computing threats and blockchain-based logging to ensure authenticity and traceability of diagnostic outputs. Evaluation across multi-center kidney ultrasound datasets and simulated adversarial/anomalous operating conditions shows that the framework achieves 96.8% diagnostic accuracy, 94.0% robustness under attack, and 94.2% anomaly detection rate, with inference latency of 412.3 ms on edge-class hardware (Jetson Xavier), meeting real-time clinical screening requirements. These results indicate that KArAt-KidneyNet can support secure and interpretable kidney disease classification in distributed consumer healthcare environments.

Keywords

anomaly detection, federated learning, healthcare IoT, Kolmogorov-Arnold attention, post-quantum cryptography, Proactive threat hunting, Soft Actor-Critic, wearable devices

Publication Title

IEEE Transactions on Consumer Electronics

ISSN

0098-3063

DOI

10.1109/TCE.2026.3680956

Volume

72

Issue

3

First Page

7105

Last Page

7120

Publisher

Institute of Electrical and Electronics Engineers

Share

COinS