基于多模态交叉注意力机制融合的1型糖尿病血糖浓度预测方法

doi:10.6040/j.issn.1671-7554.0.2025.0511

摘要/Abstract

摘要： 目的通过融合扫描式葡萄糖监测(flash glucose monitoring, FGM)数据与结构化电子健康记录(electronic health records, EHR)中的多模态信息,构建1型糖尿病(type 1 diabetes mellitus, T1DM)患者血糖浓度预测模型,以解决传统单一模态模型在捕捉复杂血糖波动规律时的局限性,为临床制定个性化控糖策略及早期风险预警提供数据支持。方法基于T1DiabetesGranada数据集,整合FGM数据、生化检测指标、人口统计学信息及诊断编码等模态特征,构建多模态时序预测模型XCLA-Net。该模型采用一维卷积神经网络提取血糖序列的局部动态特征,结合长短时记忆网络建模长期时间依赖关系,通过交叉注意力机制实现多模态语义对齐,并引入自归一化神经网络增强融合特征的稳定性。结果 XCLA-Net在RMSE、MAE、MAPE等评估指标上显著优于多种对比模型。在未来1 h和3 h预测任务中,MAPE分别为19.64%和37.81%,表明模型在不同时间尺度下具备较强的预测准确性;克拉克误差网格分析表明,模型预测结果高度集中于A区,具备良好的临床一致性。消融实验验证了交叉注意力机制、一维卷积神经网络与长短时记忆网络对提升模型预测性能的关键作用。结论本研究提出的XCLA-Net通过多模态数据融合与时序建模,显著提升了血糖预测的精度与稳定性,具备良好的临床解释性和实用价值,为T1DM患者的个性化血糖管理及早期风险预警提供有效支持。

关键词: 1型糖尿病, 多模态数据融合, 交叉注意力机制, 血糖预测, 时序建模

Abstract: Objective To develop a blood glucose concentration prediction model for patients with type 1 diabetes mellitus(T1DM)by integrating multimodal information from flash glucose monitoring(FGM)data and structured electronic health records(EHR), so as to address the limitations of traditional unimodal models in capturing complex glucose fluctuation patterns and provide data support for personalized glycemic control strategies and early risk warning in clinical practice. Methods Based on the T1DiabetesGranada dataset, the study integrated multimodal features including FGM data, biochemical test indicators, demographic information, and diagnostic codes to construct a multimodal temporal prediction model, XCLA-Net. The model employed a one-dimensional convolutional neural network(1D-CNN)to extract short-term dynamic features of the glucose sequence, combined a long short-term memory(LSTM)network to capture long-term temporal dependencies, incorporated a cross-attention mechanism for multimodal semantic alignment, and introduced a self-normalizing neural network(SNN)to enhance the stability of the fused representations. Results XCLA-Net significantly outperformed multiple baseline models in terms of error metrics such as root mean square error(RMSE), mean absolute error(MAE), and mean absolute percentage error(MAPE). The MAPE values for 1-hour and 3-hour prediction tasks were 19.64% and 37.81%, respectively, indicating strong predictive accuracy across different temporal scales. Clarke error grid analysis showed that the majority of prediction points fell within Zone A, reflecting good clinical consistency. Ablation experiments confirm the critical roles of the cross-attention mechanism, 1D-CNN, and LSTM in enhancing model performance. Conclusion The proposed XCLA-Net model effectively improves the accuracy and stability of blood glucose prediction through multimodal data fusion and temporal modeling. It demonstrates favorable clinical interpretability and practical value, providing reliable support for personalized glycemic management and early risk prediction in patients with T1DM.

Key words: Type 1 diabetes mellitus, Multimodal data fusion, Cross-attention mechanism, Blood glucose prediction, Time series modeling

中图分类号:

R587.1

王梦星,薛付忠,杨帆. 基于多模态交叉注意力机制融合的1型糖尿病血糖浓度预测方法[J]. 山东大学学报 (医学版), 2025, 63(8): 41-50.

WANG Mengxing, XUE Fuzhong, YANG Fan. Blood glucose concentration prediction method for type 1 diabetes mellitus based on multi-modal cross-attention mechanism fusion[J]. Journal of Shandong University (Health Sciences), 2025, 63(8): 41-50.

参考文献

[1] Ilonen J, Lempainen J, Veijola R. The heterogeneous pathogenesis of type 1 diabetes mellitus[J]. Nat Rev Endocrinol, 2019, 15(11): 635-650.
[2] International Diabetes Federation. IDF diabetes atlas, 11th edn[R]. Brussels, Belgium: International Diabetes Federation, 2025.
[3] Martín-Timón I, Del Cañizo-Gómez FJ. Mechanisms of hypoglycemia unawareness and implications in diabetic patients[J]. World J Diabetes, 2015, 6(7): 912-926.
[4] 师瑞, 冯磊, 唐灵通, 等. 糖尿病患者血糖波动评价指标研究进展[J]. 中华全科医学, 2022, 20(12): 2105-2109. SHI Rui, FENG Lei, TANG Lingtong, et al. Research progress on evaluation indicators of blood glucose fluctuation in patients with diabetes[J]. Chinese Journal of General Practice, 2022, 20(12): 2105-2109.
[5] Nemat H, Khadem H, Elliott J, et al. Data-driven blood glucose level prediction in type 1 diabetes: a comprehensive comparative analysis[J]. Sci Rep, 2024, 14: 21863. doi:10.1038/s41598-024-70277-x
[6] Wang Q, Molenaar P, Harsh S, et al. Personalized state-space modeling of glucose dynamics for type 1 diabetes using continuously monitored glucose, insulin dose, and meal intake: an extended Kalman filter approach[J]. J Diabetes Sci Technol, 2014, 8(2): 331-345.
[7] 肖泽秋, 李勇, 王霞. 基于PBI-CLA模型的糖尿病患者血糖浓度预测[J]. 计算机工程, 2025: 1-11. doi: 10.19678/j.issn.1000-3428.0070527 XIAO Zeqiu, LI Yong, WANG Xia. Prediction of blood glucose concentration in diabetic patients based on PBI-CLA model[J]. China Industrial Economics, 2025: 1-11. doi: 10.19678/j.issn.1000-3428.0070527
[8] 童梦, 丁国荣, 余楠, 等. 一种基于VMD-PSO-LSTM的血糖预测方法[J]. 计算机与数字工程, 2023, 51(6): 1439-1443. TONG Meng, DING Guorong, YU Nan, et al. Prediction method of blood glucose based on VMD-PSO-LSTM[J]. Computer & Digital Engineering, 2023, 51(6): 1439-1443.
[9] Wang L, Pan ZL, Liu W, et al. A dual-attention based coupling network for diabetes classification with heterogeneous data[J]. J Biomed Inform, 2023, 139: 104300. doi:10.1016/j.jbi.2023.104300
[10] Rodriguez-Leon C, Aviles-Perez MD, Banos O, et al. T1DiabetesGranada: a longitudinal multi-modal dataset of type 1 diabetes mellitus[J]. Sci Data, 2023, 10(1): 916. doi:10.1038/s41597-023-02737-4
[11] Engels JM, Diehr P. Imputation of missing longitudinal data: a comparison of methods[J]. J Clin Epidemiol, 2003, 56(10): 968-976.
[12] Dahouda MK, Joe I. A deep-learned embedding technique for categorical features encoding[J]. IEEE Access, 2021, 9: 114381-114391. doi:10.1109/ACCESS.2021.3104357
[13] Jain YK, Bhandare SK. Min max normalization based data perturbation method for privacy protection[J]. Int J Comput Commun Technol, 2013: 233-238. doi:10.47893/ijcct.2013.1201
[14] Palangi H, Deng L, Shen YL, et al. Deep sentence embedding using long short-term memory networks: analysis and application to information retrieval[J]. IEEE/ACM Trans Audio Speech Lang Process, 2016, 24(4): 694-707.
[15] Shi J, Telesca D, Suchard MA. Statistical learning techniques applied to epidemiology: a simulated case-control comparison study with logistic regression[J]. BMC Bioinformatics, 2011, 12: 37. doi:10.1186/1471-2105-12-37
[16] Zhang J, He XH, Liu Y, et al. Multi-modal cross-attention network for Alzheimers disease diagnosis with multi-modality data[J]. Comput Biol Med, 2023, 162: 107050. doi:10.1016/j.compbiomed.2023.107050
[17] Klambauer G, Unterthiner T, Mayr A, et al. Self-normalizing neural networks[EB/OL].(2017-09-07)[2025-04-26]. https://doi.org/10.48550/arXiv.1706.02515
[18] Vaswani A, Shazeer N, Parmar N, et al. Attention is all you need[EB/OL].(2023-08-02)[2025-04-26]. https://doi.org/10.48550/arXiv.1706.03762
[19] Santurkar S, Tsipras D, Ilyas A, et al. How does batch normalization help optimization?[EB/OL].(2019-04-15)[2025-04-26]. https://doi.org/10.48550/arXiv.1805.11604
[20] Srivastava N, Hinton G, Krizhevsky A, et al. Dropout: a simple way to prevent neural networks from overfitting[J]. J Mach Learn Res, 2014, 15: 1929-1958. doi: 10.5555/2627435.2670313
[21] 陶蔚, 陇盛, 刘鑫, 等. 深度学习步长自适应动量优化方法研究综述[J]. 小型微型计算机系统, 2025, 46(2): 257-265. TAO Wei, LONG Sheng, LIU Xin, et al. Review of adaptive stepsize momentum optimization methods in deep learning[J]. Journal of Chinese Computer Systems, 2025, 46(2): 257-265.
[22] Felizardo V, Garcia NM, Pombo N, et al. Data-based algorithms and models using diabetics real data for blood glucose and hypoglycaemia prediction-a systematic literature review[J]. Artif Intell Med, 2021, 118: 102120. doi:10.1016/j.artmed.2021.102120
[23] Clarke WL. The original Clarke error grid analysis(EGA)[J]. Diabetes Technol Ther, 2005, 7(5): 776-779.
[24] Shiri FM, Perumal T, Mustapha N, et al. A comprehensive overview and comparative analysis on deep learning models: CNN, RNN, LSTM, GRU[EB/OL].(2025-03-17)[2025-04-26]. https://doi.org/10.48550/arXiv.2305.17473
[25] Kamyshanska H, Memisevic R. The potential energy of an autoencoder[J]. IEEE Trans Pattern Anal Mach Intell, 2015, 37(6): 1261-1273.
[26] Deng YX, Lu L, Aponte L, et al. Deep transfer learning and data augmentation improve glucose levels prediction in type 2 diabetes patients[J]. NPJ Digit Med, 2021, 4(1): 109. doi:10.1038/s41746-021-00480-x
[27] Yang T, Yu X, Tao R, et al. Blood glucose prediction for type 2 diabetes using clustering-based domain adaptation[J]. Biomed Signal Process Contr, 2025, 105: 107629. doi:10.1016/j.bspc.2025.107629
[28] Pfützner A, Klonoff DC, Pardo S, et al. Technical aspects of the Parkes error grid[J]. J Diabetes Sci Technol, 2013, 7(5): 1275-1281.
[29] Yu X, Yang Z, Wang XZ, et al. A prior-knowledge-guided dynamic attention mechanism to predict nocturnal hypoglycemic events in type 1 diabetes[J]. BMC Med Inform Decis Mak, 2024, 24(1): 378. doi:10.1186/s12911-024-02761-3
[30] Van den Broeck G, Lykov A, Schleich M, et al. On the tractability of SHAP explanations[J]. Jair, 2022, 74: 851-886. doi:10.1613/jair.1.13283

多维度评价

Viewed

Full text

Abstract

Cited

Shared

Discussed