سامانه همایش‌ها TSTA | ثبت‌نام و ارسال مقاله کنفرانس‌ها و کنگره‌های ملی

صفحه اصلی / سی و دومین کنفرانس ملی و دهمین کنفرانس بین المللی مهندسی زیست پزشکی ایران

Robust Glucose Level Classification from NIR-Based PPG Using Morphological Features

نویسندگان :

Arian Mesforoosh Mashhad¹ Yeganeh Binafar² Mohammad Reza Akbarzadeh Totonchi³

1- دانشگاه فردوسی مشهد 2- دانشگاه فردوسی مشهد 3- دانشگاه فردوسی مشهد

کلمات کلیدی :

Diabetes classification،Photoplethysmography (PPG)،Near-infrared spectroscopy (NIRS)،Biomedical signal processing،Morphological features،Machine learning

چکیده :

Diabetes is a primary global health concern, and noninvasive monitoring could be critical for its early detection and management. This study presents a noninvasive approach to blood glucose classification using photoplethysmography (PPG) signal and machine learning approaches. However, PPG signals are biological signals that, similar to their counterparts, suffer from considerable environmental noise and patient-to-patient variability. Here, we propose a morphology-based framework for robust PPG-based Glucose classification. For this purpose, a custom-designed optical finger sensor operating at 940 nm was used to record two independent 30 s signals from fasting participants, including both healthy and diabetic subjects. After excluding low-quality signals, the final dataset included 159 subjects. Signals also underwent multi-stage filtering, normalization, and cycle-based template-matching quality control before feature extraction. We then employed the proposed framework to identify consistent cycle-shape patterns within each acquisition and verify their stability across repeated recordings. Two feature sets were compared including the cycle-based morphological and global signal-based features. Correlation analysis showed that morphology-based features were more robust and reproducible, while global signal features were less reliable under short-duration acquisitions. Multiple classifiers were tested, with Gradient Boosting achieving the highest accuracy (93.75%) using morphological features, compared to 84.38% with non-morphological features. These findings suggest that morphology-based signal analysis provides robust and salient features from short PPG signals, enabling practical and accurate noninvasive diabetes screening.