Marta Botella

Predicting Glycemia in Diabetic Patients By Evolutionary Computation and Continuous Glucose Monitoring

Diabetes mellitus is a disease that affects more than three hundreds million people worldwide. Maintaining a good control of the disease is critical to avoid not only severe long-term complications but also dangerous short-term situations. Diabetics need to decide the appropriate insulin injection, thus they need to be able to estimate the level of glucose they are going to have after a meal. In this paper we use machine learning techniques for predicting glycemia in diabetic patients. The algorithms utilize data collected from real patients by a continuous glucose monitoring system, the estimated number of carbohydrates, and insulin administration for each meal. We compare (1) non-linear regression with fixed model structure, (2) identification of prognosis models by symbolic regression using genetic programming, (3) prognosis by k-nearest-neighbor time series search, and (4) identification of prediction models by grammatical evolution. We consider predictions horizons of 30, 60, 90 and 120 minutes.

Clarke and parkes error grid analysis of diabetic glucose models obtained with evolutionary computation

Diabetes mellitus is a disease that affects to hundreds of millions of people worldwide. Maintaining a good control of the disease is critical to avoid severe long-term complications. In recent years, a lot of research has been made to improve the quality of life of the diabetic patient, especially in the automation of glucose level control. One of the main problems that arises in the (semi) automatic control of diabetes, is to obtain a model that explains the behavior of blood glucose levels with insulin, food intakes and other external factors, fitting the characteristics of each individual or patient. Recently, Grammatical Evolution (GE), has been proposed to solve this lack of models. A proposal based on GE was able to obtain customized models of five in-silico patient data with a mean percentage average error of 13.69%, modeling well also both hyper and hypoglycemic situations. In this paper we have extended the study of Error Grid Analysis (EGA) to prediction models in up to 8 in-silico patients. EGA is commonly used in Endocrinology to test the clinical significance of differences between measurements and real value of blood glucose, but has not been used before as a metric in obtention of glycemia models.

Enhancing Grammatical Evolution Through Data Augmentation: Application to Blood Glucose Forecasting

Currently, Diabetes Mellitus Type 1 patients are waiting hopefully for the arrival of the Artificial Pancreas (AP) in a near future. AP systems will control the blood glucose of people that suffer the disease, improving their lives and reducing the risks they face everyday. At the core of the AP, an algorithm will forecast future glucose levels and estimate insulin bolus sizes. Grammatical Evolution (GE) has been proved as a suitable algorithm for predicting glucose levels. Nevertheless, one the main obstacles that researches have found for training the GE models is the lack of significant amounts of data. As in many other fields in medicine, the collection of data from real patients is very complex. In this paper, we propose a data augmentation algorithm that generates synthetic glucose time series from real data. The synthetic time series can be used to train a unique GE model or to produce several GE models that work together in a combining system. Our experimental results show that, in a scarce data context, Grammatical Evolution models can get more accurate and robust predictions using data augmentation.

Data-Based Identification of Prediction Models for Glucose

Diabetes mellitus is a disease that affects to hundreds of million of people worldwide. Maintaining a good control of the disease is critical to avoid severe long-term complications. One of the main problems that arise in the (semi) automatic control of diabetes, is to get a model explaining how glucose levels in blood vary with insulin, food intakes and other factors, fitting the characteristics of each individual or patient. In this paper we compare genetic programming techniques with a set of classical identification techniques: classical simple exponential smoothing, Holts smoothing (linear, exponential and damped), classical Holt and Winters methods and auto regressive integrated moving average modeling. We consider predictions horizons of 30, 60, 90 and 120 minutes. Experimental results shows the difficulty of predicting glucose values for more than 60 minutes and the necessity of adapt GP techniques for those dynamic environments.

Forecasting glucose levels in patients with diabetes mellitus using semantic grammatical evolution and symbolic aggregate approximation

Type 1 Diabetes Mellitus can only be treated injecting insulin and glucagon into the blood stream. This research is motivated by the challenge to accurately predict future blood glucose levels of a diabetic patient so that an automatic system could decide when is necessary the injection of a bolus of insulin to keep blood sugar in the healthy range. In this paper, we have studied different evolutionary strategies based on geometric semantic genetic programming and grammatical evolution. The main contribution of this paper is the use of the symbolic aggregate approximation representation of the glucose time series that allow us to define easily semantic operators. We have developed a new strategy that combines grammatical evolution with the geometric semantic approach and that, thanks to the use of the symbolic representation, improves the previous models of glucose time series. We also present a variation of this technique that employs a univariate marginal distribution algorithm to tune the parameters of the symbolic representation. The experimental results are compared against classical grammatical evolution and geometric semantic hill climbing genetic programming. The baseline is provided by the conventional ARIMA model. Our experimental results show that the symbolic representation improves the performance of the geometric semantic strategy and reduces the number of mistakes that, if in an automatic system, would put patients health at risk.

Data augmentation and evolutionary algorithms to improve the prediction of blood glucose levels in scarcity of training data

Diabetes Mellitus Type 1 patients are waiting for the arrival of the Artificial Pancreas. Artificial Pancreas systems will control the blood glucose of patients, improving their quality of life and reducing the risks they face daily. At the core of the Artificial Pancreas, an algorithm will forecast future glucose levels and estimate insulin bolus sizes. Grammatical Evolution has been proved as a suitable algorithm for predicting glucose levels. Nevertheless, one of the main obstacles that researches have found for training the Grammatical Evolution models is the lack of significant amounts of data. As in many other fields in medicine, the collection of data from real patients is very complex along with the fact that the patients response can vary in a high degree due to a lot of personal factors which can be seen as different scenarios. In this paper, we propose both a classification system for scenario selection and a data augmentation algorithm that generates synthetic glucose time series from real data. Our experimental results show that, in a scarce data context, Grammatical Evolution models can get more accurate and robust predictions using scenario selection and data augmentation.

A genetic algorithm approach to customizing a glucose model based on usual therapeutic parameters

Type 1 diabetes mellitus is a chronic disease characterized by the increase of glucose in the blood due to a defect in the action or in the production of insulin. For completely autonomous glycemic regulation, a model would be required which permits the future evolution of blood glucose to be estimated. One of the main problems in identifying models is the high variability of glucose profiles both from one patient to another, and in the same patient under not very different conditions. In this paper, we propose a method using an evolutionary algorithm to define the values of the parameters of a minimal model based on standard clinical therapy for a several-day horizon. The algorithm is able to show the trend of blood glucose in a 5-day profile by adjusting the glucose model.