Scholarlink

The increasing dependence on IoT-enabled insulin pumps improves diabetes treatment while posing cybersecurity threats, including unauthorized insulin manipulation. Current security protocols prioritize the protection of data transmission yet inadequately identify harmful insulin injections at the device level. This work presents a TinyML-based anomaly detection system for the real-time diagnosis of irregular insulin dosages through the analysis of patient-specific glucose-insulin patterns. Our method utilizes lightweight, on-device intelligence within a Blockchain-secured healthcare architecture to mitigate pre-transmission vulnerabilities. Utilizing simulated data, we illustrate the system’s capability to identify malicious overdoses and underdoses with an accuracy of 98.39% and 93.14%, respectively. For real-world relevance, clinical validation and adaptive learning will follow.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Securing Insulin Pumps Against Malicious Injections: A TinyML-Based Anomaly Detection Approach

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Khadija Tlemçani,
  • Kebira Azbeg,
  • Leila Fetjah,
  • Ouail Ouchetto,
  • Said Jai Andaloussi

摘要

The increasing dependence on IoT-enabled insulin pumps improves diabetes treatment while posing cybersecurity threats, including unauthorized insulin manipulation. Current security protocols prioritize the protection of data transmission yet inadequately identify harmful insulin injections at the device level. This work presents a TinyML-based anomaly detection system for the real-time diagnosis of irregular insulin dosages through the analysis of patient-specific glucose-insulin patterns. Our method utilizes lightweight, on-device intelligence within a Blockchain-secured healthcare architecture to mitigate pre-transmission vulnerabilities. Utilizing simulated data, we illustrate the system’s capability to identify malicious overdoses and underdoses with an accuracy of 98.39% and 93.14%, respectively. For real-world relevance, clinical validation and adaptive learning will follow.