Nocturnal low glucose detection in healthy elderly from one-lead ECG using convolutional denoising autoencoders. (September 2020)

• Record Type: Journal Article
• Authors: Porumb, Mihaela; Griffen, Corbin; Hattersley, John; Pecchia, Leandro
• Is Part Of: Biomedical signal processing and control. Volume 62(2020)
• Publication Date: 2020-09
• Subjects: Nocturnal hypoglycemia -- Convolutional denoising autoencoder -- Convolutional neural networks; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2020.102054
• Languages: English
• ISSNs: 1746-8094

Abstract

Abstract: Hypoglycemia is a condition caused by low blood glucose levels, mainly affecting people with diabetes. Low levels of blood glucose can be dangerous, causing multiple complications over time. Hypoglycemia affects the repolarization characteristics of the heart. This study presents a personalized, deep learning-based approach that enables nocturnal hypoglycemia detection using raw electrocardiogram (ECG) signals, recorded with non-invasive, wearable devices. The study was carried out in a calorimeter room during 24–36 h monitoring on twenty-five, healthy, elderly participants. The results of this study (accuracy of ≈ 90%) provide evidence for the feasibility of a non-invasive, ECG-based hypoglycemia alarming system. Moreover, leveraging an unsupervised method (i.e. convolutional denoising autoencoder) and visualizing the ECG heartbeats in the embeddings space, clear heartbeats clusters grouped by glucose levels could be determined, showing that specific patterns in the input heartbeats are indeed glucose discriminative and the autoencoder could successfully capture those patterns.