AttenGluco: Multimodal Transformer-Based Blood Glucose Forecasting on AI-READI Dataset

TL;DR

AttenGluco introduces a multimodal Transformer framework for long-term blood glucose forecasting by fusing CGM data with activity signals through cross-attention and capturing long-range patterns via multi-scale attention. The model omits a decoder and uses embedding/positional encoding to output an $m$-step CGM forecast, evaluated on the AI-READI dataset across four cohorts. Empirical results show AttenGluco reduces RMSE by roughly 10–15% and MAE by similar margins, with robust performance across prediction horizons (5, 30, 60 minutes) and improved correlation, indicating potential for precision diabetes management. The study also analyzes training regimes, including isolated subject training and cohort-wise fine-tuning, and assesses forgetting under continual learning, highlighting both the benefits and challenges of adapting the model to new cohorts in real-world deployments.

Abstract

Diabetes is a chronic metabolic disorder characterized by persistently high blood glucose levels (BGLs), leading to severe complications such as cardiovascular disease, neuropathy, and retinopathy. Predicting BGLs enables patients to maintain glucose levels within a safe range and allows caregivers to take proactive measures through lifestyle modifications. Continuous Glucose Monitoring (CGM) systems provide real-time tracking, offering a valuable tool for monitoring BGLs. However, accurately forecasting BGLs remains challenging due to fluctuations due to physical activity, diet, and other factors. Recent deep learning models show promise in improving BGL prediction. Nonetheless, forecasting BGLs accurately from multimodal, irregularly sampled data over long prediction horizons remains a challenging research problem. In this paper, we propose AttenGluco, a multimodal Transformer-based framework for long-term blood glucose prediction. AttenGluco employs cross-attention to effectively integrate CGM and activity data, addressing challenges in fusing data with different sampling rates. Moreover, it employs multi-scale attention to capture long-term dependencies in temporal data, enhancing forecasting accuracy. To evaluate the performance of AttenGluco, we conduct forecasting experiments on the recently released AIREADI dataset, analyzing its predictive accuracy across different subject cohorts including healthy individuals, people with prediabetes, and those with type 2 diabetes. Furthermore, we investigate its performance improvements and forgetting behavior as new cohorts are introduced. Our evaluations show that AttenGluco improves all error metrics, such as root mean square error (RMSE), mean absolute error (MAE), and correlation, compared to the multimodal LSTM model. AttenGluco outperforms this baseline model by about 10% and 15% in terms of RMSE and MAE, respectively.

Figures (4)

• Figure 1: Overview of the AttenGluco framework including sensing module, data preparation, and forecasting model.
• Figure 2: AttenGluco model architecture consists of cross-attention and multi-scale attention to forecast BGL.
• Figure 3: The RMSE of each subject in (a) the baseline model and (b) AttenGluco under the cohort-wise scenario. As observed, AttenGluco's RMSE decreases significantly within each cohort as more subjects are incorporated into the fine-tuning process.
• Figure 4: Fine-tuning the model on new cohorts leads to the loss of knowledge from previous ones.