Fault and meal detection by redundant continuous glucose monitors and the unscented Kalman filter

An ensemble machine learning approach for the detection of unannounced meals to enhance postprandial glucose control

BACKGROUND

Hybrid automated insulin delivery systems enhance postprandial glucose control in type 1 diabetes, however, meal announcements are burdensome. To overcome this, we propose a machine learning-based automated meal detection approach; METHODS:: A heterogeneous ensemble method combining an artificial neural network, random forest, and logistic regression was employed. Trained and tested on data from two in-silico cohorts comprising 20 and 47 patients. It accounted for various meal sizes (moderate to high) and glucose appearance rates (slow and rapid absorbing). To produce an optimal prediction model, three ensemble configurations were used: logical AND, majority voting, and logical OR. In addition to the in-silico data, the proposed meal detector was also trained and tested using the OhioT1DM dataset. Finally, the meal detector is combined with a bolus insulin compensation scheme; RESULTS:: The ensemble majority voting obtained the best meal detector results for both the in-silico and OhioT1DM cohorts with a sensitivity of 77%, 94%, 61%, precision of 96%, 89%, 72%, F1-score of 85%, 91%, 66%, and with false positives per day values of 0.05, 0.19, 0.17, respectively. Automatic meal detection with insulin compensation has been performed in open-loop insulin therapy using the AND ensemble, chosen for its lower false positive rate. Time-in-range has significantly increased 10.48% and 16.03%, time above range was reduced by 5.16% and 11.85%, with a minimal time below range increase of 0.35% and 2.69% for both in-silico cohorts, respectively, compared to the results without a meal detector; CONCLUSION:: To increase the overall accuracy and robustness of the predictions, this ensemble methodology aims to take advantage of each base model's strengths. All of the results point to the potential application of the proposed meal detector as a separate module for the detection of meals in automated insulin delivery systems to achieve improved glycemic control.

A Deep Learning Framework for Automatic Meal Detection and Estimation in Artificial Pancreas Systems

Current artificial pancreas (AP) systems are hybrid closed-loop systems that require manual meal announcements to manage postprandial glucose control effectively. This poses a cognitive burden and challenge to users with T1D since this relies on frequent user engagement to maintain tight glucose control. In order to move towards fully automated closed-loop glucose control, we propose an algorithm based on a deep learning framework that performs multitask quantile regression, for both meal detection and carbohydrate estimation. Our proposed method is evaluated in silico on 10 adult subjects from the UVa/Padova simulator with a Bio-inspired Artificial Pancreas (BiAP) control algorithm over a 2 month period. Three different configurations of the AP are evaluated -BiAP without meal announcement (BiAP-NMA), BiAP with meal announcement (BiAP-MA), and BiAP with meal detection (BiAP-MD). We present results showing an improvement of BiAP-MD over BiAP-NMA, demonstrating 144.5 ± 6.8 mg/dL mean blood glucose level (-4.4 mg/dL, p< 0.01) and 77.8 ± 6.3% mean time between 70 and 180 mg/dL (+3.9%, p< 0.001). This improvement in control is realised without a significant increase in mean in hypoglycaemia (+0.1%, p= 0.4). In terms of detection of meals and snacks, the proposed method on average achieves 93% precision and 76% recall with a detection delay time of 38 ± 15 min (92% precision, 92% recall, and 37 min detection time for meals only). Furthermore, BiAP-MD handles hypoglycaemia better than BiAP-MA based on CVGA assessment with fewer control errors (10% vs. 20%). This study suggests that multitask quantile regression can improve the capability of AP systems for postprandial glucose control without increasing hypoglycaemia.

Fault Tolerant Strategies for Automated Insulin Delivery Considering the Human Component: Current and Future Perspectives

Automated Insulin Delivery (AID) are systems developed for daily use by people with type 1 diabetes (T1D). To ensure the safety of users, it is essential to consider how the human factor affects the performance and safety of these devices. While there are numerous publications on hardware-related failures of AID systems, there are few studies on the human component of the system. From a control point of view, people with T1D using AID systems are at the same time the plant to be controlled and the plant operator. Therefore, users may induce faults in the controller, sensors, actuators, and the plant itself. Strategies to cope with the human interaction in AID systems are needed for further development of the technology. In this paper, we present an analysis of potential faults introduced by AID users when the system is under normal operation. This is followed by a review of current fault tolerant control (FTC) approaches to identify missing areas of research. The paper concludes with a discussion on future directions for the new generation of FTC AID systems.

A Hybrid Automata Approach for Monitoring the Patient in the Loop in Artificial Pancreas Systems

The use of automated insulin delivery systems has become a reality for people with type 1 diabetes (T1D), with several hybrid systems already on the market. One of the particularities of this technology is that the patient is in the loop. People with T1D are the plant to control and also a plant operator, because they may have to provide information to the control loop. The most immediate information provided by patients that affects performance and safety are the announcement of meals and exercise. Therefore, to ensure safety and performance, the human factor impact needs to be addressed by designing fault monitoring strategies. In this paper, a monitoring system is developed to diagnose potential patient modes and faults. The monitoring system is based on the residual generation of a bank of observers. To that aim, a linear parameter varying (LPV) polytopic representation of the system is adopted and a bank of Kalman filters is designed using linear matrix inequalities (LMI). The system uncertainty is propagated using a zonotopic-set representation, which allows determining confidence bounds for each of the observer outputs and residuals. For the detection of modes, a hybrid automaton model is generated and diagnosis is performed by interpreting the events and transitions within the automaton. The developed system is tested in simulation, showing the potential benefits of using the proposed approach for artificial pancreas systems.

Toward an Optimal Definition of Hypoglycemia with Continuous Glucose Monitoring

BACKGROUND AND OBJECTIVE

As continuous glucose monitoring (CGM) becomes common in research and clinical practice, there is a need to understand how CGM-based hypoglycemia relates to hypoglycemia episodes defined conventionally as patient reported hypoglycemia (PRH). Data show that CGM identify many episodes of low interstitial glucose (LIG) that are not experienced by patients, and so the aim of this study is to use different PRH simulations to optimize CGM parameters of threshold (h) and duration (d) to provide the best PRH detection performance.

METHODS

The algorithm uses particle Markov chain Monte Carlo optimization to identify the optimal h and d which maximize an objective function for detecting PRH. We tested our algorithm by creating three different cases of PRH simulations.

RESULTS

We added three types of simulated PRH events to 10 weeks of anonymized CGM data from 96 type 1 diabetes people to see if the algorithm can detect the optimal parameters set out in the simulations. In simulation 1, we changed the locations of PRHs with respect to LIG episodes in the CGM signal to simulate random optimal LIG parameters for every individual. In simulation 2, the PRHs are CGM glucose <3.9 mmol/L followed by at least 20 min of rise > 0.11 mmol/L/min. Simulation 3 is like simulation 2 but with glucose threshold of 3.0 mmol/L. The median [interquartile range] of deviation between the optimized (found by the algorithm) and the optimal (known) h and d are -0.07% [-0.4, 1.9] and -1.3% [-5.9, 6.8], respectively across the subjects for simulation 1. The mean [min max] of the optimized LIG parameters are h = 3.8 [3.7, 3.8] mmol/L and d = 12 [10, 14] min for simulation 2 and they are h = 3.0 [2.9, 3] mmol/L and d = 10 [8, 14] min for simulation 3 across a 10-fold cross validation.

CONCLUSIONS

This work demonstrates the feasibility of the algorithm to find the best-fit definition of CGM-based hypoglycemia for PRH detection. In a prospective clinical study collecting CGM and PRH, the current algorithm will be used to optimize the definition of hypoglycemia with respect to PRH with the ambition of using the resulted definition as a surrogate for PRH in clinical practice.

Artificial Pancreas: Evaluating the ARG Algorithm Without Meal Announcement

BACKGROUND

Either under standard basal-bolus treatment or hybrid closed-loop control, subjects with type 1 diabetes are required to count carbohydrates (CHOs). However, CHO counting is not only burdensome but also prone to errors. Recently, an artificial pancreas algorithm that does not require premeal insulin boluses-the so-called automatic regulation of glucose (ARG)-was introduced. In its first pilot clinical study, although the exact CHO counting was not required, subjects still needed to announce the meal time and classify the meal size.

METHOD

An automatic switching signal generator (SSG) is proposed in this work to remove the manual mealtime announcement from the control strategy. The SSG is based on a Kalman filter and works with continuous glucose monitoring readings only.

RESULTS

The ARG algorithm with unannounced meals (ARG_um) was tested in silico under the effect of different types of mixed meals and intrapatient variability, and contrasted with the ARG algorithm with announced meals (ARG_am). Simulations reveal that, for slow-absorbing meals, the time in the euglycemic range, [70-180] mg/dL, increases using the unannounced strategy (ARG_am: 78.1 [68.6-80.2]% (median [IQR]) and ARG_um: 87.8 [84.5-90.6]%), while similar results were found with fast-absorbing meals (ARG_am: 87.4 [86.0-88.9]% and ARG_um: 87.6 [86.1-88.8]%). On the other hand, when intrapatient variability is considered, time in euglycemia is also comparable (ARG_am: 81.4 [75.4-83.5]% and ARG_um: 80.9 [77.0-85.1]%).

CONCLUSION

In silico results indicate that it is feasible to perform an in vivo evaluation of the ARG algorithm with unannounced meals.

Feasibility of Early Meal Detection Based on Abdominal Sound

In classical approaches for an artificial pancreas, continuous glucose monitoring (CGM) is the only measured variable used for insulin dosing and additional control functions. The CGM values are subject to time delays and slow dynamics between blood and the sensing location. These time lags compromise the controller's performance in maintaining (near to) normal glucose levels. Meal information could enhance the control outcome. However, meal announcement by the user is not reliable, and it takes 30 min to 40 min from meal onset until a meal is detected by methods based on CGM. In this pilot study, the use of bowel sounds for meal detection was investigated. In particular, we focused on whether bowel sounds change qualitatively during or shortly after meal ingestion. After fasting for at least 4 h, 11 healthy volunteers ingested a lunch meal at their usual time. Abdominal sound was recorded by a condenser microphone that was attached to the right upper quadrant of the abdomen by medical tape. Features that describe the power distribution over the frequency spectrum were extracted and used for classification by support vector machines. These classifiers were trained in a leave-one-out cross-validation scheme. Meals could be detected on average 10 min (std: 4.4 min) after they had started. Half of these were detected without false alarms. This shows that abdominal sound monitoring could provide an early meal detection. Further studies should investigate this possibility on a larger population in more general settings.

Pattern Recognition Reveals Characteristic Postprandial Glucose Changes: Non-Individualized Meal Detection in Diabetes Mellitus Type 1

Accurate continuous glucose monitoring (CGM) is essential for fully automated glucose control in diabetes mellitus type 1. State-of-the-art glucose control systems automatically regulate the basal insulin infusion. Users still need to manually announce meals to dose the prandial insulin boluses. An automated meal detection could release the user and improve the glucose regulation. In this study, patterns in the postprandial CGM data are exploited for meal detection. Binary classifiers are trained to recognize the postprandial pattern in horizons of the estimated glucose rate of appearance and in CGM data. The appearance rate is determined by moving horizon estimation based on a simple model. Linear discriminant analysis (LDA) is used for classification. The proposed method is compared to methods that detect meals when thresholds are violated. Diabetes care data from 12 free-living pediatric patients was downloaded during regular screening. Experts identified meals and their start by retrospective evaluation. The classification was tested by cross-validation. Compared to the threshold-based methods, LDA showed higher sensitivity to meals with a low rate of false alarms. Classifying horizons outperformed the other methods also with respect to time of detection. The onset of meals can be detected by pattern recognition based on estimated model states and consecutive CGM measurements. No individual tuning is necessary. This makes the method easily adopted in the clinical practice.

Unannounced Meals in the Artificial Pancreas: Detection Using Continuous Glucose Monitoring

The artificial pancreas (AP) system is designed to regulate blood glucose in subjects with type 1 diabetes using a continuous glucose monitor informed controller that adjusts insulin infusion via an insulin pump. However, current AP developments are mainly hybrid closed-loop systems that include feed-forward actions triggered by the announcement of meals or exercise. The first step to fully closing the loop in the AP requires removing meal announcement, which is currently the most effective way to alleviate postprandial hyperglycemia due to the delay in insulin action. Here, a novel approach to meal detection in the AP is presented using a sliding window and computing the normalized cross-covariance between measured glucose and the forward difference of a disturbance term, estimated from an augmented minimal model using an Unscented Kalman Filter. Three different tunings were applied to the same meal detection algorithm: (1) a high sensitivity tuning, (2) a trade-off tuning that has a high amount of meals detected and a low amount of false positives (FP), and (3) a low FP tuning. For the three tunings sensitivities 99 ± 2%, 93 ± 5%, and 47 ± 12% were achieved, respectively. A sensitivity analysis was also performed and found that higher carbohydrate quantities and faster rates of glucose appearance result in favorable meal detection outcomes.