{"title":"Hybrid Mamba-MoE model for non-invasive blood glucose prediction","authors":"Ümit Şentürk , Gökhan Adıgüzel , Kemal Polat","doi":"10.1016/j.compeleceng.2025.110549","DOIUrl":null,"url":null,"abstract":"<div><div>Diabetes is a growing global health concern, necessitating accurate and non-invasive blood glucose monitoring solutions. Traditional glucose measurement methods, such as finger-prick tests and continuous glucose monitoring (CGM) systems, pose challenges related to patient adherence, discomfort, and cost, highlighting the need for advanced AI-driven alternatives. This study proposes a hybrid Mamba-MoE model, integrating State Space Models (SSM) and Mixture of Experts (MoE) architectures, to enhance the accuracy, robustness, and real-time feasibility of Photoplethysmography (PPG)-based blood glucose prediction. The methodology involves PPG data acquisition, preprocessing, synthetic signal generation, and model optimization. Wavelet denoising and moving average smoothing techniques were employed to mitigate motion artifacts and noise. At the same time, segmentation, normalization, and temporal feature extraction enhanced the quality and consistency of the PPG signals. A synthetic PPG signal augmentation framework was implemented to address data scarcity and imbalance, utilizing temporal epoch shifting and regression-based glucose interpolation. The Mamba-MoE model was trained and validated using a 5-fold cross-validation approach, ensuring robust performance across diverse patient profiles. Experimental results demonstrate that Mamba-MoE significantly outperforms traditional machine learning and deep learning models, achieving Zone A accuracy of 98.16% and a low RMSE of 7.99, making it a strong candidate for real-time, wearable glucose monitoring applications. The study underscores the potential of AI-driven hybrid models in improving non-invasive diabetes management while addressing clinical reliability, scalability, and computational efficiency. Future research should focus on expanding dataset diversity, integrating multimodal physiological signals, and optimizing computational efficiency for real-world deployment. By advancing AI-powered non-invasive glucose monitoring, this study paves the way for a more accessible, accurate, and patient-friendly approach to diabetes management.</div></div>","PeriodicalId":50630,"journal":{"name":"Computers & Electrical Engineering","volume":"127 ","pages":"Article 110549"},"PeriodicalIF":4.0000,"publicationDate":"2025-07-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Computers & Electrical Engineering","FirstCategoryId":"94","ListUrlMain":"https://www.sciencedirect.com/science/article/pii/S0045790625004926","RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q1","JCRName":"COMPUTER SCIENCE, HARDWARE & ARCHITECTURE","Score":null,"Total":0}
引用次数: 0
Abstract
Diabetes is a growing global health concern, necessitating accurate and non-invasive blood glucose monitoring solutions. Traditional glucose measurement methods, such as finger-prick tests and continuous glucose monitoring (CGM) systems, pose challenges related to patient adherence, discomfort, and cost, highlighting the need for advanced AI-driven alternatives. This study proposes a hybrid Mamba-MoE model, integrating State Space Models (SSM) and Mixture of Experts (MoE) architectures, to enhance the accuracy, robustness, and real-time feasibility of Photoplethysmography (PPG)-based blood glucose prediction. The methodology involves PPG data acquisition, preprocessing, synthetic signal generation, and model optimization. Wavelet denoising and moving average smoothing techniques were employed to mitigate motion artifacts and noise. At the same time, segmentation, normalization, and temporal feature extraction enhanced the quality and consistency of the PPG signals. A synthetic PPG signal augmentation framework was implemented to address data scarcity and imbalance, utilizing temporal epoch shifting and regression-based glucose interpolation. The Mamba-MoE model was trained and validated using a 5-fold cross-validation approach, ensuring robust performance across diverse patient profiles. Experimental results demonstrate that Mamba-MoE significantly outperforms traditional machine learning and deep learning models, achieving Zone A accuracy of 98.16% and a low RMSE of 7.99, making it a strong candidate for real-time, wearable glucose monitoring applications. The study underscores the potential of AI-driven hybrid models in improving non-invasive diabetes management while addressing clinical reliability, scalability, and computational efficiency. Future research should focus on expanding dataset diversity, integrating multimodal physiological signals, and optimizing computational efficiency for real-world deployment. By advancing AI-powered non-invasive glucose monitoring, this study paves the way for a more accessible, accurate, and patient-friendly approach to diabetes management.
期刊介绍:
The impact of computers has nowhere been more revolutionary than in electrical engineering. The design, analysis, and operation of electrical and electronic systems are now dominated by computers, a transformation that has been motivated by the natural ease of interface between computers and electrical systems, and the promise of spectacular improvements in speed and efficiency.
Published since 1973, Computers & Electrical Engineering provides rapid publication of topical research into the integration of computer technology and computational techniques with electrical and electronic systems. The journal publishes papers featuring novel implementations of computers and computational techniques in areas like signal and image processing, high-performance computing, parallel processing, and communications. Special attention will be paid to papers describing innovative architectures, algorithms, and software tools.