Key result
Fine-tuned foundation models with dietary context improve postprandial CGM forecasting over baseline methods.
Why the study?
Accurate forecasting of continuous glucose monitoring trajectories can support proactive diabetes management, but the utility of time-series foundation models and added multimodal dietary context was unproven.
Does adding multimodal dietary context to fine-tuned time-series foundation models improve postprandial continuous glucose monitoring forecasting in individuals with and without diabetes?
Does adding multimodal dietary context to fine-tuned time-series foundation models improve postprandial continuous glucose monitoring forecasting in individuals with and without diabetes?
Fine-tuning foundation models and incorporating multimodal dietary context significantly improves the accuracy of postprandial continuous glucose monitoring forecasting.
May aid postprandial glucose prediction in diabetes care; leaves open prospective validation of multimodal foundation models.
Introduction and Objective: Accurate forecasting of continuous glucose monitoring (CGM) trajectories can support proactive diabetes management. We evaluated time-series foundation models (FMs) for CGM forecasting in two populations: type 1 diabetes and a combined non-diabetes/type 2 diabetes group. We assessed whether adding multimodal dietary context improves predictive performance. Methods: Using 8 CGM datasets, we compared FMs (zero-shot and fine-tuned) with baseline methods across multiple forecast horizons and input history lengths. To test dietary effects, we augmented CGM inputs in the CGMacros dataset with (1) food-image embeddings generated by a ConvNeXT model fine-tuned on Food-101 and (2) structured macronutrient features. Models were evaluated on held-out test sets and postprandial subset. Results: In zero-shot settings, FMs provided little improvement over baseline models. After task-specific fine-tuning, FMs achieved lower prediction error across horizons. Adding meal information improved postprandial forecasting vs CGM-only models. Food-image embeddings were most beneficial for visually distinctive meals, whereas macronutrient features were more robust for visually complex or ambiguous meals (Fig). Conclusion: FMs require task-specific fine-tuning to improve CGM forecasting. Dietary context improves post-meal prediction, with images and macronutrients contributing complementary value for meal-aware forecasting. Disclosure B. Zhang: None. K. Zhou: None. H. Cheng: None. H. Yang: None. J. Zhou: None. H. Zhou: None. Funding National Institutes of Health under grants R35 GM141798 (HZ), R01 HG006139 (HZ and JZ), and R01 DK142026 (HZ and JZ); National Science Foundation under grants DMS 2054253 (HZ and JZ) and IIS 2205441 (HZ and JZ).
No takes yet. Share an insight, caveat, or question.
ZHANG et al. (2026) studied Type 1 diabetes and non-diabetes/type 2 diabetes. Fine-tuned foundation models with multimodal dietary context vs. Baseline models and zero-shot foundation models was evaluated on Prediction error across forecast horizons. Fine-tuned foundation models augmented with multimodal dietary context (food images and macronutrients) improved postprandial continuous glucose monitoring forecasting compared to baseline models.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: