Blood Glucose Prediction in Type 1 Diabetes Using Deep Learning on the Edge

Abstract

Real-time blood glucose (BG) prediction can enhance decision support systems for insulin dosing such as bolus calculators and closed-loop systems for insulin delivery. Deep learning has been proven to achieve state-of-the-art performance in BG prediction. However, it is usually seen as a very computationally expensive approach, hence difficult to implement in wearable medical devices such as transmitters in continuous glucose monitoring (CGM) systems. In this work, we introduce a novel deep learning framework to predict BG levels with the edge inference on a microcontroller unit embedded in a lowpower system. By using glucose measurements from a CGM sensor and a recurrent neural network that builds on long-short term memory, the personalized models achieves state-of-the-art performance on a clinical data set obtained from 12 subjects with T1D. In particular, the proposed method achieves an average root mean square error of 19.10 ± 2.04 for a 30-minute prediction horizon (PH) and 32.61 ± 3.45 for a 60-minute PH with high clinical accuracy. Notably, the framework has been optimized to achieve a minimum use of hardware resources (34KB FLASH and 1KB SRAM) as well as an execution time of 22 ms for low power operations (8 µW). The presented system has the potential to be implemented in wearable medical devices for diabetes care (CGM and insulin pumps) and to be integrated within an Internet of Things platform.Index Terms-Diabetes, deep learning, artificial pancreas, Internet of things, microcontroller, edge inference, LSTM.

Embedded Machine Learning Using Microcontrollers in Wearable and Ambulatory Systems for Health and Care Applications: A Review

“…The other tools are more specific, targeting a certain family of MCU; such as CMSIS-NN for ARM based MCUs [58], [84], [85], while PULP-NN and NEMO for PULP based MCUs [45], [89], [90]. More specific targeted MCU software tools are X-CUBE-AI for use with only STM32 MCUs [45], [59], [82], [86]- [88], [91], GAPflow for GAP MCUs, and DORY with current support for only GAP8. Due to these compatibility concerns, the choice of MCU and software tools was co-dependent.…”

“…Edge inference was also proposed in [59] in the context of a wearable artificial pancreas systems for patients with Type 1 Diabetes (T1D). The system input readings, acquired from a continuous glucose monitoring (CGM) sensor, were used to predict blood glucose through the use of an RNN model, based on LSTM layers.…”

Type 1 Diabetes Hypoglycemia Prediction Algorithms: Systematic Review

“…These patients strictly follow the given research guidelines or are in a monitoring environment, which abstains from everyday life where patients mostly do not monitor events, such as heart rate, regularly, which are usually essential for these methodologies. We also noticed that several methodologies used a limited number of features [7,9,11,13,18,19]. This can have a significant impact on the final results, as several factors can affect blood glucose levels, each with different severity.…”

“…Another deep learning framework for predicting blood glucose levels was recently developed [19], which used edge inference on a microcontroller unit. The performance of the models was evaluated based on a clinical data set acquired from 12 patients with T1D whose glucose was measured with a CGM, as well as through a long short-term memory artificial recurrent neural network.…”

A Multi-Network Fusion Prediction Method for Patient Blood Glucose Concentration Prediction and Hyper/Hypoglycemia Blood Glucose Warning

“…Zhu et al [7] processed BG measurements from CGM sensors for real-time prediction of BG by building a long and short-term memory recurrent neural network. In addition, many researchers have explored studies on BG prediction using only historical CGM data, starting from regression models, artificial neural networks, and kernel-based methods [8][9][10].…”