A Federated Learning-Inspired Evolutionary Algorithm: Application to Glucose Prediction.

In this paper, we propose an innovative Federated Learning-inspired evolutionary framework. Its main novelty is that this is the first time that an Evolutionary Algorithm is employed on its own to directly perform Federated Learning activity. A further novelty resides in the fact that, differently from the other Federated Learning frameworks in the literature, ours can efficiently deal at the same time with two relevant issues in Machine Learning, i.e., data privacy and interpretability of the solutions. Our framework consists of a master/slave approach in which each slave contains local data, protecting sensible private data, and exploits an evolutionary algorithm to generate prediction models. The master shares through the slaves the locally learned models that emerge on each slave. Sharing these local models results in global models. Being that data privacy and interpretability are very significant in the medical domain, the algorithm is tested to forecast future glucose values for diabetic patients by exploiting a Grammatical Evolution algorithm. The effectiveness of this knowledge-sharing process is assessed experimentally by comparing the proposed framework with another where no exchange of local models occurs. The results show that the performance of the proposed approach is better and demonstrate the validity of its sharing process for the emergence of local models for personal diabetes management, usable as efficient global models. When further subjects not involved in the learning process are considered, the models discovered by our framework show higher generalization capability than those achieved without knowledge sharing: the improvement provided by knowledge sharing is equal to about 3.03% for precision, 1.56% for recall, 3.17% for F1, and 1.56% for accuracy. Moreover, statistical analysis reveals the statistical superiority of model exchange with respect to the case of no exchange taking place.

Distributed Assessment of Virtual Insulin-Pump Settings Using SmartCGMS and DMMS.R for Diabetes Treatment.

Diabetes is a heterogeneous group of diseases that share a common trait of elevated blood glucose levels. Insulin lowers this level by promoting glucose utilization, thus avoiding short- and long-term organ damage due to the elevated blood glucose level. A patient with diabetes uses an insulin pump to dose insulin. The pump uses a controller to compute and dose the correct amount of insulin to keep blood glucose levels in a safe range. Insulin-pump controller development is an ongoing process aiming at fully closed-loop control. Controllers entering the market must be evaluated for safety. We propose an evaluation method that exploits an FDA-approved diabetic patient simulator. The method evaluates a Cartesian product of individual insulin-pump parameters with a fine degree of granularity. As this is a computationally intensive task, the simulator executes on a distributed cluster. We identify safe and risky combinations of insulin-pump parameter settings by applying the binomial model and decision tree to this product. As a result, we obtain a tool for insulin-pump settings and controller safety assessment. In this paper, we demonstrate the tool with the Low-Glucose Suspend and OpenAPS controllers. For average ± standard deviation, LGS and OpenAPS exhibited 1.7 ± 0.6% and 3.2 ± 1.8% of local extrema (i.e., good insulin-pump settings) out of all the entire Cartesian products, respectively. A continuous region around the best-discovered settings (i.e., the global extremum) of the insulin-pump settings spread across 4.0 ± 1.1% and 4.1 ± 1.3% of the Cartesian products, respectively.

Physiological reconstruction of blood glucose level using CGMS-signals only.

Patient with diabetes must regularly monitor blood glucose level. Drawing a blood sample is a painful and discomfort experience. Alternatively, the patient measures interstitial fluid glucose level with a sensor installed in subcutaneous tissue. Then, a model of glucose dynamics calculates blood glucose level from the sensor-measured, i.e., interstitial fluid glucose level of subcutaneous tissue. Interstitial fluid glucose level can significantly differ from blood glucose level. The sensor is either factory-calibrated, or the patient calibrates the sensor periodically by drawing blood samples, when glucose levels of both compartments are steady. In both cases, the sensor lifetime is limited up to 14 days. This is the present state of the art. With a physiological model, we would like to prolong the sensor lifetime with an adaptive approach, while requiring no additional blood sample. Prolonging sensor's lifetime, while reducing the associated discomfort, would considerably improve patient's quality of life. We demonstrate that it is possible to determine personalized model parameters from multiple CGMS-signals only, using an animal experiment with a hyperglycemic clamp. The experimenter injected separate glucose and insulin boluses to trigger rapid changes, on which we evaluated the ability to react to non-steady glucose levels in different compartments. With the proposed model, 70%, 80% and 95% of the calculated blood glucose levels had relative error less than or equal to 21.9%, 32.5% and 43.6% respectively. Without the model, accuracy of the sensor-estimated blood glucose level decreased to 39.4%, 49.9% and 99.0% relative errors. This confirms feasibility of the proposed method.

Predicting glucose level with an adapted branch predictor.

BACKGROUND AND OBJECTIVE: Diabetes mellitus manifests as prolonged elevated blood glucose levels resulting from impaired insulin production. Such high glucose levels over a long period of time damage multiple internal organs. To mitigate this condition, researchers and engineers have developed the closed loop artificial pancreas consisting of a continuous glucose monitor and an insulin pump connected via a microcontroller or smartphone. A problem, however, is how to accurately predict short term future glucose levels in order to exert efficient glucose-level control. Much work in the literature focuses on least prediction error as a key metric and therefore pursues complex prediction methods such a deep learning. Such an approach neglects other important and significant design issues such as method complexity (impacting interpretability and safety), hardware requirements for low-power devices such as the insulin pump, the required amount of input data for training (potentially rendering the method infeasible for new patients), and the fact that very small improvements in accuracy may not have significant clinical benefit. METHODS: We propose a novel low-complexity, explainable blood glucose prediction method derived from the Intel P6 branch predictor algorithm. We use Meta-Differential Evolution to determine predictor parameters on training data splits of the benchmark datasets we use. A comparison is made between our new algorithm and a state-of-the-art deep-learning method for blood glucose level prediction. RESULTS: To evaluate the new method, the Blood Glucose Level Prediction Challenge benchmark dataset is utilised. On the official test data split after training, the state-of-the-art deep learning method predicted glucose levels 30 min ahead of current time with 96.3% of predicted glucose levels having relative error less than 30% (which is equivalent to the safe zone of the Surveillance Error Grid). Our simpler, interpretable approach prolonged the prediction horizon by another 5 min with 95.8% of predicted glucose levels of all patients having relative error less than 30%. CONCLUSIONS: When considering predictive performance as assessed using the Blood Glucose Level Prediction Challenge benchmark dataset and Surveillance Error Grid metrics, we found that the new algorithm delivered comparable predictive accuracy performance, while operating only on the glucose-level signal with considerably less computational complexity.

Using meta-differential evolution to enhance a calculation of a continuous blood glucose level.

We developed a new model of glucose dynamics. The model calculates blood glucose level as a function of transcapillary glucose transport. In previous studies, we validated the model with animal experiments. We used analytical method to determine model parameters. In this study, we validate the model with subjects with type 1 diabetes. In addition, we combine the analytic method with meta-differential evolution. To validate the model with human patients, we obtained a data set of type 1 diabetes study that was coordinated by Jaeb Center for Health Research. We calculated a continuous blood glucose level from continuously measured interstitial fluid glucose level. We used 6 different scenarios to ensure robust validation of the calculation. Over 96% of calculated blood glucose levels fit A+B zones of the Clarke Error Grid. No data set required any correction of model parameters during the time course of measuring. We successfully verified the possibility of calculating a continuous blood glucose level of subjects with type 1 diabetes. This study signals a successful transition of our research from an animal experiment to a human patient. Researchers can test our model with their data on-line at https://diabetes.zcu.cz.

Blood glucose level reconstruction as a function of transcapillary glucose transport.

A diabetic patient occasionally undergoes a detailed monitoring of their glucose levels. Over the course of a few days, a monitoring system provides a detailed track of their interstitial fluid glucose levels measured in their subcutaneous tissue. A discrepancy in the blood and interstitial fluid glucose levels is unimportant because the blood glucose levels are not measured continuously. Approximately five blood glucose level samples are taken per day, and the interstitial fluid glucose level is usually measured every 5min. An increased frequency of blood glucose level sampling would cause discomfort for the patient; thus, there is a need for methods to estimate blood glucose levels from the glucose levels measured in subcutaneous tissue. The Steil-Rebrin model is widely used to describe the relationship between blood and interstitial fluid glucose dynamics. However, we measured glucose level patterns for which the Steil-Rebrin model does not hold. Therefore, we based our research on a different model that relates present blood and interstitial fluid glucose levels to future interstitial fluid glucose levels. Using this model, we derived an improved model for calculating blood glucose levels. In the experiments conducted, this model outperformed the Steil-Rebrin model while introducing no additional requirements for glucose sample collection. In subcutaneous tissue, 26.71% of the calculated blood glucose levels had absolute values of relative differences from smoothed measured blood glucose levels less than or equal to 5% using the Steil-Rebrin model. However, the same difference interval was encountered in 63.01% of the calculated blood glucose levels using the proposed model. In addition, 79.45% of the levels calculated with the Steil-Rebrin model compared with 95.21% of the levels calculated with the proposed model had 20% difference intervals.

Glucose predictability, blood capillary permeability, and glucose utilization rate in subcutaneous, skeletal muscle, and visceral fat tissues.

This study suggests an approach for the comparison and evaluation of particular compartments with modest experimental setup costs. A glucose level prediction model was used to evaluate the compartment's glucose transport rate across the blood capillary membrane and the glucose utilization rate by the cells. The glucose levels of the blood, subcutaneous tissue, skeletal muscle tissue, and visceral fat were obtained in experiments conducted on hereditary hypertriglyceridemic rats. After the blood glucose level had undergone a rapid change, the experimenter attempted to reach a steady blood glucose level by manually correcting the glucose infusion rate and maintaining a constant insulin infusion rate. The interstitial fluid glucose levels of subcutaneous tissue, skeletal muscle tissue, and visceral fat were evaluated to determine the reaction delay compared with the change in the blood glucose level, the interstitial fluid glucose level predictability, the blood capillary permeability, the effect of the concentration gradient, and the glucose utilization rate. Based on these data, the glucose transport rate across the capillary membrane and the utilization rate in a particular tissue were determined. The rates obtained were successfully verified against positron emission tomography experiments. The subcutaneous tissue exhibits the lowest and the most predictable glucose utilization rate, whereas the skeletal muscle tissue has the greatest glucose utilization rate. In contrast, the visceral fat is the least predictable and has the shortest reaction delay compared with the change in the blood glucose level. The reaction delays obtained for the subcutaneous tissue and skeletal muscle tissue were found to be approximately equal using a metric based on the time required to reach half of the increase in the interstitial fluid glucose level.

Prediction of interstitial glucose level.

Glucose is an important source of energy for cells. In clinical practice, we measure glucose level in blood and interstitial fluid. Each method has its pros and cons, and both levels correlate with each other. As the body tries to maintain the glucose level within a particular range to avoid adverse effects, it is desirable to predict future glucose levels in order to aid provided health care. We can see this desire in research, e.g., research on glucose transporters of cells. As yet another example, we can see it with diabetic patients, patients in a metabolic intensive care unit, particularly. In this paper, a glucose level prediction method is proposed.

Estimating reaction delay for glucose level prediction.

The pancreas, liver and hypothalamus have a regulatory function in the glucose homeostasis. As the blood glucose level changes, these compartments react and the level changes again. Subsequently to this reaction, the interstitial glucose level changes with some delay. In this paper, I propose a hypothesis that the change of the blood glucose level includes information about the estimated rate with which the hypothalamus expects the blood glucose level to return to normal range, by means of regulatory mechanisms of glucose homeostasis. As the interstitial glucose level change reflects the blood glucose level change, I propose a method to estimate the blood-to-interstitial glucose level delay. It is an important factor for glucose level prediction. Once the delay was calculated, it was possible to relate the present blood glucose level and future interstitial glucose level with such coefficients, which do not seem to change over the time of the experiment to a significant extent. Perhaps, it is a parameterization of regulatory processes of glucose homeostasis, which could be possibly encoded within hypothalamus set-points. The delays were constant per subject and ranged from 7 min up to 34 min for hereditary hypertriglyceridemic rats of 230-480 g weight, in experiments with a variable glucose infusion rate.