Background: Blood glucose (BG) management is crucial for type-1 diabetes patients resulting in the necessity of reliable artificial pancreas or insulin infusion systems. In recent years, deep learning techniques have been utilized for a more accurate BG level prediction system. However, continuous glucose monitoring (CGM) readings are susceptible to sensor errors. As a result, inaccurate CGM readings would affect BG prediction and make it unreliable, even if the most optimal machine learning model is used. Methods: In this work, we propose a novel approach to predicting blood glucose level with a stacked Long short-term memory (LSTM) based deep recurrent neural network (RNN) model considering sensor fault. We use the Kalman smoothing technique for the correction of the inaccurate CGM readings due to sensor error. Results: For the OhioT1DM (2018) dataset, containing eight weeks’ data from six different patients, we achieve an average RMSE of 6.45 and 17.24 mg/dl for 30 min and 60 min of prediction horizon (PH), respectively. Conclusions: To the best of our knowledge, this is the leading average prediction accuracy for the ohioT1DM dataset. Different physiological information, e.g., Kalman smoothed CGM data, carbohydrates from the meal, bolus insulin, and cumulative step counts in a fixed time interval, are crafted to represent meaningful features used as input to the model. The goal of our approach is to lower the difference between the predicted CGM values and the fingerstick blood glucose readings—the ground truth. Our results indicate that the proposed approach is feasible for more reliable BG forecasting that might improve the performance of the artificial pancreas and insulin infusion system for T1D diabetes management.

Medical Cyber-Physical Systems (MCPS) hold the promise of reducing human errors and optimizing healthcare by delivering new ways to monitor, diagnose and treat patients through integrated clinical environments (ICE). Despite the benefits provided by MCPS, many of the ICE medical devices have not been designed to satisfy cybersecurity requirements and, consequently, are vulnerable to recent attacks. Nowadays, ransomware attacks account for 85% of all malware in healthcare, and more than 70% of attacks confirmed data disclosure. With the goal of improving this situation, the main contribution of this paper is an automatic, intelligent and real-time system to detect, classify, and mitigate ransomware in ICE. The proposed solution is fully integrated with the ICE++ architecture, our previous work, and makes use of Machine Learning (ML) techniques to detect and classify the spreading phase of ransomware attacks affecting ICE. Additionally, Network Function Virtualization (NFV) and Software Defined Networking (SDN)paradigms are considered to mitigate the ransomware spreading by isolating and replacing infected devices. Different experiments returned a precision/recall of 92.32%/99.97% in anomaly detection, an accuracy of 99.99% in ransomware classification, and promising detection and mitigation times. Finally, different labelled ransomware datasets in ICE have been created and made publicly available.

In modern hospitals, patients are treated using a wide array of medical devices that are increasingly interacting with each other over the network, thus offering a perfect example of a cyber-physical system. We study the safety of a medical device system for the physiologic closed-loop control of drug infusion. The main contribution of the paper is the verification approach for the safety properties of closed-loop medical device systems. We demonstrate, using a case study, that the approach can be applied to a system of clinical importance. Our method combines simulation-based analysis of a detailed model of the system that contains continuous patient dynamics with model checking of a more abstract timed automata model. We show that the relationship between the two models preserves the crucial aspect of the timing behavior that ensures the conservativeness of the safety analysis. We also describe system design that can provide open-loop safety under network failure.

Conformal prediction is a statistical tool for producing prediction regions of machine learning models that are valid with high probability. However, applying conformal prediction to time series data leads to conservative prediction regions. In fact, to obtain prediction regions over T time steps with confidence 1--delta, previous works require that each individual prediction region is valid with confidence 1--delta/T. We propose an optimization-based method for reducing this conservatism to enable long horizon planning and verification when using learning-enabled time series predictors. Instead of considering prediction errors individually at each time step, we consider a parameterized prediction error over multiple time steps. By optimizing the parameters over an additional dataset, we find prediction regions that are not conservative. We show that this problem can be cast as a mixed integer linear complementarity program (MILCP), which we then relax into a linear complementarity program (LCP). Additionally, we prove that the relaxed LP has the same optimal cost as the original MILCP. Finally, we demonstrate the efficacy of our method on case studies using pedestrian trajectory predictors and F16 fighter jet altitude predictors.