Logo: to the web site of Uppsala University

Type 1 Diabetes (T1D) has challenged humanity for over 3,500 years, from its earliest descriptions in ancient medical texts to today’s cutting-edge biotechnological solutions. Even today, T1D remains a growing global health concern and is the second most common chronic disease among children in Sweden. Although there is no cure, significant progress has been made in treatment strategies, particularly through the development of artificial pancreas (AP) systems. An AP is a closed-loop insulin delivery system that integrates a glucose sensor, an insulin pump, and a control algorithm to mimic the glucose-regulating function of a healthy pancreas. By continuously adjusting insulin infusion based on real-time glucose measurements, AP systems reduce the burden of diabetes management and improve long-term health outcomes.

However, as AP systems rely on sensor data and wireless communication, they are susceptible to cyber threats. One such threat is sensor deception attacks, where an attacker manipulates the sensor readings to mislead the insulin delivery algorithm, potentially causing excessively low or high glucose levels. Detecting such attacks is particularly challenging due to natural glucose fluctuations caused by meal intake, which can mask adversarial manipulation.

The need for computationally efficient and reliable anomaly detection algorithms is paramount, particularly in wearable medical devices such as the AP. To this end, model-based anomaly detection schemes offer a mathematically rigorous and lightweight alternative, enabling timely and accurate detection of anomalies, including cyberattacks, while meeting the real-time constraints of safety-critical systems. This thesis aims to advance model-based detection methods by integrating residual generation and evaluation techniques, optimizing the trade-off between detection speed and false alarm minimization. By contributing to the development of secure AP systems, this research aims to enhance patient safety and improve the quality of life for individuals with T1D.