Identification of Physical Activity Type in People with Diabetes: A Spectrogram-based Approach

Abstract

Background: Individuals with Type 1 Diabetes (T1D) require close glucose monitoringto prevent both short- and long-term complications. Physical activity (PA) is a significantsource of variability in metabolic dynamics, leading to glycemic fluctuations that depend onthe type, intensity, and duration of the exercise. Accurately monitoring and classifying thetype of PA is crucial for optimizing glycemic control and minimizing the risk of hypoglycemia.Methods: This study utilizes the largest clinical trial of PA in people with T1D to date,the Type 1 Diabetes and Exercise Initiative (T1DEXI), which included both structured andunstructured PA sessions, to develop an online classification approach for identifying thetype of PA (aerobic, interval, resistance). A computationally efficient Convolutional NeuralNetwork (CNN) was trained on time-frequency representations (spectrograms) of step countand heart rate signals, readily available from wearable devices, from the structured PAsessions of the T1DEXI dataset. The proposed methodology presents an ad-hoc processfor designing the spectrograms based on the CNN architecture to optimize the classifier’sperformance.Results: The CNN-based classification approach was implemented using spectrogramsof 5 and 30-minute signals, resulting in two classifiers that achieve high classification accuracywhen evaluated on the structured PA sessions. The 5-minute classifier was then applied tounstructured PA sessions, where the predicted distribution of glucose changes for the activitytypes was consistent with clinical evidence.Conclusion: These results demonstrate the potential of the proposed approach for itsintegration into decision support systems or automated insulin delivery systems, enablingimproved glucose management during exercise in T1D.