Detection of Meals and Physical Activity Events From Free-Living Data of People With Diabetes

Abstract

Background: Predicting carbohydrate intake and physical activity in people with diabetes is crucial for improving blood glucose concentration regulation. Patterns of individual behavior can be detected from historical free-living data to predict meal and exercise times. Data collected in free-living may have missing values and forgotten manual entries. While machine learning (ML) can capture meal and exercise times, missing values, noise, and errors in data can reduce the accuracy of ML algorithms. Methods: Two recurrent neural networks (RNNs) are developed with original and imputed data sets to assess detection accuracy of meal and exercise events. Continuous glucose monitoring (CGM) data, insulin infused from pump data, and manual meal and exercise entries from free-living data are used to predict meals, exercise, and their concurrent occurrence. They contain missing values of various lengths in time, noise, and outliers. Results: The accuracy of RNN models range from 89.9% to 95.7% for identifying the state of event (meal, exercise, both, or neither) for various users. “No meal or exercise” state is determined with 94.58% accuracy by using the best RNN (long short-term memory [LSTM] with 1D Convolution). Detection accuracy with this RNN is 98.05% for meals, 93.42% for exercise, and 55.56% for concurrent meal-exercise events. Conclusions: The meal and exercise times detected by the RNN models can be used to warn people for entering meal and exercise information to hybrid closed-loop automated insulin delivery systems. Reliable accuracy for event detection necessitates powerful ML and large data sets. The use of additional sensors and algorithms for detecting these events and their characteristics provides a more accurate alternative.

Exploring Nutritional Influence on Blood Glucose Forecasting for Type 1 Diabetes Using Explainable AI

“…Another aspect that should be addressed relates the input features used for the prediction of postprandial glycemic levels: in fact, these have been selected based on theoretical and physiological considerations [23], [61], and available data, but may not capture all relevant variables. Indeed, beyond nutritional factors, it would be of considerable interest to explore the influence of variables associated with physical activity and the psychological wellbeing of patients [73], [74]. In this regard, XAI-based features impact findings could be used to carry out feature selection to improve model performance by identifying the most influential input features [75]- [77].…”

Artificial Intelligence and Machine Learning for Improving Glycemic Control in Diabetes: Best Practices, Pitfalls, and Opportunities

“…Deep learning methods have been successfully reported for automated detection of diabetic retinopathy and diabetic macular edema in retinal fundus photographs [23,24]. Closed-loop control algorithms for automated insulin and other hormone delivery have been augmented with ML methods for automating the detection of hypoglycemia during exercise [16,18,[25][26][27], meal detection [28][29][30][31][32][33][34][35][36][37] and time series prediction models that can be incorporated into model predictive control algorithms to achieve over 70% time in glucose target range (TIR, 70-180 mg/dL) [16,26]. Anomaly detection techniques can identify disturbances and complications in diabetes management [38,39].…”

“…The Tidepool data set has been used to train ML algorithms for predicting overnight hypoglycemia at the time when a person goes to sleep [55] and also to predict short-term glucose and hypoglycemia up to 60 minutes in the future [26]. It has also been used to develop a DNN to detect meals, exercise and their concurrent occurrences as well [25,56].…”

Physical Activity Detection for Diabetes Mellitus Patients Using Recurrent Neural Networks

“…However, conventional methods often fail to provide comprehensive insights, prompting a search for innovative solutions. In this pursuit, recurrent neural networks (RNNs) [ 26 ] have emerged as a highly promising tool to detect and analyze physical activity patterns in individuals with TDM1 [ 27 ]. RNNs, specifically designed for processing sequential data, prove to be exceptionally adept at recognizing temporal dependencies in human movement.…”