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.

Implantable Medical Devices (IMDs) have been widely used to treat chronic diseases such as cardiac arrhythmia and diabetes. Many IMDs are enabled with wireless communication capabilities and can communicate with an outside programmer/reader wirelessly. With the rapid growth of IMDs, IMD security becomes a critical issue since attacks on IMDs may directly harm the patient. Typical IMDs have very limited resource in terms of energy, computation and storage. In this research, we identify a new kind of attacks on IMDs - Resource Depletion (RD) attacks that could deplete IMD resources (e.g., battery power) quickly. The RD attacks could reduce the lifetime of an IMD from several years to a few weeks. The attacks can be easily launched but can not be defended by traditional cryptographic approaches. In this paper, we propose to utilize the patient's IMD access pattern and we design a novel Support Vector Machine (SVM) based scheme to address the RD attacks. Our SVM-based scheme is very effective in defending the RD attacks. Our experimental results show that the average detection rate of the SVM-based scheme is above 90%.

Implantable Medical Devices (IMDs) are widely used to treat chronic diseases. Nowadays, many IMDs can wirelessly communicate with an outside programmer (reader). However, the wireless access also introduces security concerns. An attacker may get an IMD reader and gain access to a patient's IMD. IMD security is an important issue since attacks on IMDs may directly harm the patient. A number of research groups have studied IMD security issues when the patient is in nonemergency situations. However, these security schemes usually require the patient's participation, and they may not work during emergencies (e.g., when the patient is in comma) for various reasons. In this paper, we propose a light-weight secure access control scheme for IMDs during emergencies. Our scheme utilizes patient's biometric information to prevent unauthorized access to IMDs. The scheme consists of two levels: level 1 employs some basic biometric information of the patient and it is lightweight; level 2 utilizes patients' iris data for authentication and it is very effective. In this research, we also make contributions in human iris verification: we discover that it is possible to perform iris verification by comparing partial iris data rather than the entire iris data. This significantly reduces the overhead of iris verification, which is critical for resource-limited IMDs. We evaluate the performance of our schemes by using real iris data sets. Our experimental results show that the secure access control scheme is very effective and has small overhead (hence feasible for IMDs). Specifically, the false acceptance rate (FAR) and false rejection rate (FRR) of our secure access control scheme are close to 0.000% with suitable threshold, and the memory and computation overheads are acceptable. Our analysis shows that the secure access control scheme reduces computation overhead by an average of 58%.

Wireless insulin pumps have been widely deployed in hospitals and home healthcare systems. Most of these insulin pump systems have limited security mechanisms embedded to protect them from malicious attacks. In this paper, two attacks against insulin pump systems via wireless links are investigated: a single acute overdose with a significant amount of medication, and chronic overdose with an insignificant amount of extra medication over a long time period, e.g., several months. These attacks can be launched unobtrusively and may jeopardize patients' lives. It is very important and urgent to protect patients from these attacks. To address this issue, we propose a novel patient infusion pattern based access control scheme (PIPAC) for wireless insulin pumps. This scheme employs a supervised learning approach to learn normal patient infusions pattern with the dosage amount, rate, and time of infusion, which are automatically recorded in insulin pump logs. The generated regression models are used to dynamically configure a safety infusion range for abnormal infusion identification. The proposed algorithm is evaluated with real insulin pump logs used by several patients for up to 6 months. The evaluation results demonstrate that our scheme can reliably detect the single overdose attack with a success rate up to 98% and defend against the chronic overdose attack with a very high success rate.

The most common eye infection in people with diabetes is diabetic retinopathy (DR). It might cause blurred vision or even total blindness. Therefore, it is essential to promote early detection to prevent or alleviate the impact of DR. However, due to the possibility that symptoms may not be noticeable in the early stages of DR, it is difficult for doctors to identify them. Therefore, numerous predictive models based on machine learning (ML) and deep learning (DL) have been developed to determine all stages of DR. However, existing DR classification models cannot classify every DR stage or use a computationally heavy approach. Common metrics such as accuracy, F1 score, precision, recall, and AUC-ROC score are not reliable for assessing DR grading. This is because they do not account for two key factors: the severity of the discrepancy between the assigned and predicted grades and the ordered nature of the DR grading scale. This research proposes computationally efficient ensemble methods for the classification of DR. These methods leverage pre-trained model weights, reducing training time and resource requirements. In addition, data augmentation techniques are used to address data limitations, improve features, and improve generalization. This combination offers a promising approach for accurate and robust DR grading. In particular, we take advantage of transfer learning using models trained on DR data and employ CLAHE for image enhancement and Gaussian blur for noise reduction. We propose a three-layer classifier that incorporates dropout and ReLU activation. This design aims to minimize overfitting while effectively extracting features and assigning DR grades. We prioritize the Quadratic Weighted Kappa (QWK) metric due to its sensitivity to label discrepancies, which is crucial for an accurate diagnosis of DR. This combined approach achieves state-of-the-art QWK scores (0.901, 0.967 and 0.944) in the Eyepacs, Aptos, and Messidor datasets.

Wireless insulin pumps have been widely deployed in hospitals and home healthcare systems. Most of them have limited security mechanisms embedded to protect them from malicious attacks. In this paper, two attacks against insulin pump systems via wireless links are investigated: a single acute overdose with a significant amount of medication and a chronic overdose with a small amount of extra medication over a long time period. They can be launched unobtrusively and may jeopardize patients' lives. It is very urgent to protect patients from these attacks. We propose a novel personalized patient infusion pattern based access control scheme (PIPAC) for wireless insulin pumps. This scheme employs supervised learning approaches to learn normal patient infusion patterns in terms of the dosage amount, rate, and time of infusion, which are automatically recorded in insulin pump logs. The generated regression models are used to dynamically configure a safe infusion range for abnormal infusion identification. This model includes two sub models for bolus (one type of insulin) abnormal dosage detection and basal abnormal rate detection. The proposed algorithms are evaluated with real insulin pump. The evaluation results demonstrate that our scheme is able to detect the two attacks with a very high success rate.